Adi Robertson | The Verge

Let’s build a children’s public internet

2026-07-13T10:15:54-04:00

An increasing number of people seem to agree the internet is terrible for children — allegedly addictive, destructive to self-esteem, possibly a portal to predators. Over the past year, several countries have started requiring stringent age verification or outright bans for minors. At the end of June in the US, the House of Representatives passed the Kids Internet and Digital Safety (KIDS) Act, the latest in a string of attempted online child safety regulations. A couple of days later, a Pew Research Center survey found over half of US respondents favored a ban on social media for anyone under 16. There’s a growing sense that the digital world is a public health crisis and something — no matter how extreme — must be done.

But while politicians chase elaborate and questionable methods of keeping kids away from the worst of the internet, another option is staring them in the face: Spend money to make it better. And, fortunately, I’ve got an idea: Levy a tax on major tech companies, and hand out that money for the construction of what we should call a children’s public internet.

First of all, what is a children’s public internet? I’m not suggesting an entirely separate service like France’s national proto-internet Minitel, but something more like the “public lane in the information superhighway” that author Ben Tarnoff proposed in Internet for the People — or, more tangibly, the 20th-century push for children’s public television. The goal would be to fund new or existing online services that meet two criteria: They primarily serve children, and they don’t operate for profit. Beyond that, the options are myriad. A few hypothetical grant recipients:

A library-run, community-moderated Mastodon instance for young users
An open-source, non-monetized version of Roblox
A website that offers ad-free, child- or teen-appropriate news and educational content
A protocol for reverse age verification — working with school systems or government agencies to verify users of a child-focused site are minors, while minimizing (to the extent it’s possible) privacy and security risks
A local newsletter or web portal that promotes nearby in-person family activities
A group of volunteer moderators for a children’s crafts forum

These services could be new or existing, led by institutions or individuals, developed and maintained by minors or adults, and accessible by small groups or anybody on the internet. The primary goal would be something almost everyone agrees is beneficial: cutting the profit motive out of kids’ and teenagers’ online lives.

Nearly every popular criticism of social media is, in some way, about its perverse economic incentives. Critics warn of allegedly predatory “dark patterns” that allow companies to hook users and scale endlessly, or invasive advertising based on even more invasive data collection, or moderation teams operated at the lowest feasible cost. However accurate any individual complaint is, years of leaks and court filings demonstrate that companies are constantly balancing user well-being with the need for engagement and profit — a stressor that’s only increasing as they divert resources toward AI.

Even well-meaning, socially constructive services exist in this ecosystem. Parenting networks are funneled onto Facebook and other platforms that are increasingly hostile to human connection or onto private sites that must support themselves with often-intrusive ads.

Plenty of people are theorizing ways to build a less vampiric internet, but at the policy level, we’ve seen basically two proposed solutions: punish companies until they change fundamental elements of their business models, or kick anybody of a vulnerable age offline. The entire field of popular internet regulation is built on taking things away from people — removing adults’ ability to engage with the internet on their own terms, while removing kids’ access to social and creative spaces while offering nothing in return.

So far, evidence suggests this supposedly “common-sense” solution is difficult to implement and deeply flawed. Australia implemented a blanket ban on teen social media use last year, but it appears to be largely ineffectual, with one study suggesting over 80 percent of kids maintain access. Age-gating systems remain simultaneously circumventable and deleterious to privacy. The US faces its own highly specific set of problems. A near-total lack of modern data privacy laws compounds the security risks of collecting age verification info. Members of Congress have questioned whether the deeply corrupt Trump administration will (or, given its gutting of federal agencies, even can) enforce internet regulations evenhandedly and in good faith.

A children’s public internet takes on the same basic problem from the opposite direction. Instead of forcibly constricting the existing online ecosystem, it would expand it with new and better alternatives. And yes, I mean objectively better, not just from a moral standpoint but a functional one. There’s never been a more opportune time for a public alternative to private tech. Small, nonprofit services can be smaller and rougher-edged than commercial ones, but they won’t be choked with ads, microtransactions, AI boondoggles, and other symptoms of an industry that simply doesn’t care if people like its products anymore.

Will some minors stick with Instagram and TikTok, or find the idea of a “kid stuff” internet uncool? Sure. But there’s a perennial appeal to discovering a new club that your parents can’t join. (Anecdotally, many kids are already attempting this on semi-private services like Discord, hampered by age-gating and beset by aggressive monetization and AI tools that young people in general hate.) And anyway, how cool or “adult” are a bunch of social networks where people are scared to say “kill” and “lesbian”?

Finally, a social network your parents can’t join

Internet law expert Eric Goldman — who authored a 2025 academic paper outlining the problems with existing child safety proposals — points out that there was an early attempt to make an internet for kids: the kids.us internet top-level domain. That domain was little-used and ultimately abandoned. But it’s an early example of people recognizing that youth-focused online spaces matter, a concept that makes far more sense these days, when the internet is central to virtually everyone’s lives — it just needs more than a bespoke web address to thrive.

Some might argue that young people simply need to get offline, period, swapping screentime for good old-fashioned grass touching. I’m not immune to this impulse — I’m a parent of a toddler who’s discovered YouTube. But even before the drastic upset of the covid-19 pandemic, many children’s offline worlds were constricting. Rebuilding these physical spaces is vital, but giving people an online alternative to Big Tech is more constructive (and, it sure seems, practically achievable) than sending them to stare at a wall.

In fact, alternative spaces to Big Tech already exist to capture kids’ attention, they’re just often even worse — places like 4chan and sketchy Telegram groups. These services are happy to flout age-gating laws and other restrictions on kids’ access while giving adult predators free rein.

And while this program would most immediately benefit children, the goal would be modeling a better internet for everyone. Open-source software could be repurposed to serve all ages, public websites could be valuable for adults as well as children, and any successful service could be cloned for adults. Similar to city-run municipal broadband, and adjacent to the public-tech movement Tarnoff has dubbed digital sewer socialism, it could provide much-needed competition that pushes tech companies to offer better service. Meanwhile, even these companies might see a nominal tax as better than onerous regulations and as an opportunity for good PR.

There are, of course, plenty of practical details to hammer out. You might need to tack on some additional requirements: a ban on commercial activity by users (similar to the stance of nonprofit fiction platform Archive of Our Own), perhaps, and a requirement that any software be open-source. There’s the matter of figuring out which government agency would oversee the program, what category of companies would be taxed, who would evaluate the grants, and how they’d be evaluated for effectiveness, for instance.

But a lot of the obvious objections seem specious. Some grant recipient or other will almost inevitably have a serious security or moderation failure, but almost certainly with less harm than its commercial counterparts. The program will be a lightning rod for complaints about government efficiency and culture war topics, but so are cancer vaccines. At a time when nothing seems to work, why not swing for the fences and try something new?

The US government helped create the internet — and for all the flaws of the digital world, it’s a damaged space that’s worth repairing. It’s time for the government to help re-create it, too.

Moves of the Diamond Hand is an unfinished, irresistibly weird dice-based RPG

2026-06-22T08:26:00-04:00

From its opening minutes, Moves of the Diamond Hand is upfront about what it offers: You’re going to have a lot of strange conversations, and you’re going to roll a lot of dice. Get on board with this proposition, and the reward is one of the most creative roleplaying games I’ve seen in years, even if its many mysteries won’t be resolved until 2027.

Moves of the Diamond Hand is an Early Access videogame available on PC, macOS, and steamOS (including the Steam Deck, where I played it) from musician and game designer Cosmo D. The game looks and feels like a 2000s-era first-person RPG or immersive sim: environments are grimy, stark, and blocky; characters’ features are stretched over smooth heads a bit too small for their faces; an eerie soundtrack pulses over all. You’ll arrive on a train and immediately meet an old mentor, disgraced by some kind of political scandal. You convey your desire to join a powerful organization called Circus X, then declare which of several wildly different paths you’ll take into its fold — you can try joining the city council, but could find equal success crafting the perfect sandwich or joining the best band.

These options help introduce the central mechanic. The game gives you one upgradeable die for each of seven stats, ranging from standard fare like Physique and Observation to the more idiosyncratic Cooking and Music. To set a challenge, it will roll a die corresponding to one of those attributes, and you’ve got to match or beat it with your own roll.

Once you emerge into the train station, the complexity quickly multiplies. There are a plethora of sub-mechanics including cooking, performing music, laundering disguises, and mixing cocktails — all of which add additional dice with unique quirks. You can selectively re-roll dice in a manner similar to Yahtzee, introducing an element of strategy within each encounter, and your final score (win or lose) is translated into experience points. The basic system was introduced in Cosmo D’s last game, Betrayal at Club Low, but in a less flexible and elegant form; Diamond Hand feels like its evolution. (Disclosure: My husband has provided outside feedback for Cosmo D’s games.)

It’s all a little intimidating at first. But the game allows you to ease into its options, which happens quickly, since you’re rolling for virtually every action and verbal exchange from making small talk to opening a door. There’s a meaningful element of chance to all this, without descending into unbounded randomness. Some rolls can be mathematically impossible to win or lose at a given skill level, but it’s possible to still damage your health or gain an unwelcome status effect with safe challenges, preventing them from becoming purely rote. You can retry most actions if you fail them, but they’ll become slightly more difficult on a second attempt, so there’s a constant balancing act of deciding when to take the initial leap. The ambient low-level risk makes even simple spaces feel substantive and engaging — it negates the common RPG urge to speed through environmental detail and flavor text while looking for the “real” parts of the game.

Through countless skill checks, you’ll internalize the odd logic of the game’s world. The setting, Off-Peak City, is a garish metropolis shaped by the machinations of sinister corporations, corrupt politicians, and shady operatives, but also musicians, restaurateurs, and literally and figuratively underground tailors — a neon retro-future for streetwise aesthetes. What might be niche skills in any other game prove extraordinarily powerful here. The Music stat, whose uses include sewing (machines can, among other options, be literally operated by improvisation), calming aggressive animal-human hybrids (by whistling tunes), and mixology (which can be performed “rhythmically”), is arguably the single strongest power in the game.

Circus X, you’ll soon learn, is a secretive arts institution that influences everything from politics to the sandwich supply chain — imagine the Factory crossed with the Freemasons. While pursuing membership, you’re embroiled in a local election between a scandal-plagued technocrat, a former boy-band star, and the corporate-controlled clone of a mayor from decades past. In place of a Maltese Falcon, everyone’s scheming for control of a sentient Big Mouth Billy Bass. And meddling behind the scenes is the mysterious, anarchic Diamond Hand, frequently alluded to but not explained.

Diamond Hand’s story evokes real-world parallels, but as a jumping-off point for something that’s rich and alive in its own right. In perhaps the most obvious example, a company in Off-Peak City is pumping the place full of clones, supplanting human artists with corporate-guardrailed regurgitations of old media. But rather than stop at commentary, the game walks this out to explore the idea that clones are also conscious beings who are frustrated by their creative limits and lack of autonomy, while letting human characters reflect on their own relationship with nostalgia and artistic taste.

Put this all together and you’ve got a hard-boiled sci-fi thriller involving subway busking, finding library books, stumping for politicians, harvesting lettuce, arguing about jazz, and doing laundry, infused with the lizard-brain appeal of a nonstop game of chance. It’s irresistible.

Most of Diamond Hand’s main quests end in roadblocks, because its Early Access build includes only the first two of six chapters, with the next scheduled for this summer and a full launch set for the spring of 2027. But even in its current state, Diamond Hand is dense and tantalizing, delivering a string of absurd premises and dry humor with a straight face. (Among many tossed-off jokes that are also actual game mechanics, local pizza-makers require everyone to bake their own pie, so if you don’t like your order, you have only yourself to blame.) You’re granted experience points for letting characters ramble through their backstories and opinions — which lands somewhere between a sly gag about RPG infodumping and a straightforwardly clever decision — but the dialogue pays off even without that prize.

And for all its dystopian elements, there’s something idealistic about a world where art, for good or ill, deeply matters. Diamond Hand may be a work in progress, but it’s a recipe for becoming obsessed with skill and perfection, chasing the world’s greatest sandwich and the string of lucky dice rolls that will get you there.

Ferrari reveals its first EV, with design help from Jony Ive

2026-05-26T10:25:01-04:00

The Ferrari Luce will start at €550,000 in Italy, but US pricing hasn’t been announced. | Image: Ferrari

After months of teasers, Ferrari is offering the first full view of its Luce electric vehicle. The Luce is notable not just for being Ferrari’s first EV, but for being designed in collaboration with Jony Ive and Mark Newson at their collective LoveFrom. It’s also going to be Ferrari’s second four-door car and its first five-seat one.

We already knew Ive and Newson were working on the Luce’s interiors, which were shown off earlier this year. Now Ferrari says LoveFrom was allowed to “define the design direction of the project from the outset,” inside and out.

Tim Stevens reporting for Engadget offers a few firsthand impressions of the car, which he describes as feeling more like an SUV than a traditional sports car. He wasn’t able to try out fully functioning interior controls or get a test drive, but he notes that with four motors and 1,035 horsepower, it ought to be set up for success. He did offer praise for the car’s sound, which apparently picks up and amplifies vibrations from the rear motors instead of being fully synthesized like many EVs. And he mentions that while there’s no US price set, it will start at €550,000 in Italy, making it the most expensive Ferrari yet.

Cox Media fined after bragging it spied on users through their phones

2026-05-25T14:28:30-04:00

An exceptionally weird controversy has come back to haunt Cox Media and a pair of marketing firms, which claimed they were secretly listening to users via phones and smart devices — despite little evidence they actually could. On Thursday the Federal Trade Commission announced that Cox, MindSift, and 1010 Digital Works would pay a total of $930,000 to settle allegations that they were in fact lying about spying on people to target ads.

As chronicled by Techdirt a couple of years ago, Cox publicly boasted about a system called Voice Data back in 2023, telling potential digital marketing clients they could ensure “every casual conversation between two consumers becomes a tool for you to target, retarget, and retain customers.” It compared the tech to an episode of Black Mirror and described it as a real version of the persistent, largely unsubstantiated rumor that social media companies routinely listen to users through phone microphones. Cox backpedaled and denied it was listening to conversations, but 404 Media published multiple internal pitch decks making essentially the same highly dystopian claim.

At the time, there were significant doubts this was actually happening, and the FTC complaints back this up. “This service did not, in fact, listen in on consumers’ conversations or use voice data at all — nor did the service accurately place ads in customers’ desired locations,” it says in its press release. “Instead, the service the companies provided consisted of reselling — at a significant markup — email lists obtained from other data brokers.” The agency also says the companies lied about consumers having opted into this system — so even if they could spy on people, it alleges, they’d still have been breaking the law.

All the evidence revealed so far in Musk v. Altman

2026-05-22T14:42:38-04:00

The Musk v. Altman trial is underway, and that means exhibits, or the evidence to be presented in court, are being revealed piece by piece. So far, email exchanges, photos, and corporate documents are circulating from the earliest days of OpenAI — and from before the AI lab even had a name. Some high-level takeaways: Nvidia CEO Jensen Huang gave OpenAI an in-demand supercomputer, Musk largely drafted OpenAI’s mission and heavily influenced its early structure, OpenAI CEO Sam Altman appeared to want to lean heavily on Y Combinator for early support for OpenAI, OpenAI president Greg Brockman and Ilya Sutskever worried about Musk’s level of control over the company, and Musk highlighted the importance of a nonprofit with a mission of broadly beneficial AI.

Musk’s buzzy lawsuit, which began its jury trial on Monday in a federal courtroom in California, names Altman, Brockman, and OpenAI investor Microsoft as defendants. The claims vary against each party and have, over the past couple of years, included breaching OpenAI’s charitable trust, fraud, and unjust enrichment. But ultimately, Musk’s lawsuit boils down to whether or not OpenAI deviated from its founding mission of ensuring that artificial general intelligence — an often vaguely defined term that denotes AI systems that equal or surpass human intelligence — benefits all of humanity. It’s the latest in a yearslong string of legal actions against OpenAI and its executives by Musk, who cofounded the AI lab alongside Altman and Brockman and was an early investor. (Musk also owns xAI, an AI lab that directly competes with OpenAI, and is owned by parent company SpaceX.)

Former OpenAI employees and people close to both companies have been watching this particular lawsuit with a close eye, since the outcome of a jury trial could have affected how OpenAI runs its business and controls its quickly advancing technology. Plus, OpenAI and SpaceX are both reportedly racing to go public this year, so they’re more in the public eye than ever.

The lawsuit discovery process had already unearthed a lot of eyebrow-raising communications between AI industry executives, from emails between Altman and Sutskever to entries from Brockman’s own diary. Even texts between Meta CEO Mark Zuckerberg and Musk were made public. But that was all before the jury trial started — now, there’s even more set to be revealed.

Here’s an exhaustive list of all the exhibits that have been made public so far and the biggest takeaways from each one. Admittedly, not every item is necessarily interesting, so we’ve flagged the most important ones with an asterisk. The Verge will keep updating the list as more are added.

Documents released April 29, 2026

*Exhibit No. 5

A June 2015 email exchange between Altman and Musk. Altman lays out a five-part plan involving an AI lab with a mission to “create the first general AI and use it for individual empowerment—ie, the distributed version of the future that seems the safest. More generally, safety should be a first-class requirement.”

He suggests that they start with seven to 10 people and expand from there, using an extra Y Combinator building located in Mountain View. Governance-wise, Altman names five people to start, proposing himself, Musk, Bill Gates, Pierre Omidyar, and Dustin Moskovitz. “The technology would be owned by the foundation and used ‘for the good of the world’, and in cases where it’s not obvious how that should be applied the 5 of us would decide,” Altman writes. He adds that the researchers working at the lab would have “significant financial upside … uncorrelated to what they build, which should eliminate some of the conflict,” and suggests paying them a “competitive salary” and awarding them equity in Y Combinator. He also says they should get someone to “run the team” but that that person “probably shouldn’t be on the governance board.”

Altman goes on to ask Musk whether he’ll be involved in the AI lab in addition to governance, potentially coming by once a month to talk about progress or at least being publicly supportive to help with recruiting. As a model, he names Peter Thiel’s “part-time partner” involvement at Y Combinator.

Finally, Altman mentions a “regulation letter,” seeming to imply that the AI lab was going to write a letter calling for AI regulation. He says he’s happy to leave Musk off as a public signatory.

Musk replies, “Agree on all.”

*Exhibit No. 7

In an October 2015 email exchange between Altman and Musk, Altman suggests starting with a $100 million commitment by Musk and asks if he could donate an additional $30 million over the next five years. He says Bill Gates isn’t yet committed to donating but that he hopes to “have him locked down next week,” adding that he believes Mark Zuckerberg likely won’t come through due to his own AI lab, Facebook AI Research (FAIR). He also suggests that he and Musk start as the first two members of the Safety Board with the potential to add three other members over the following year, calling it the “‘second key’ for releasing anything that could be dangerous.”

Musk responds, “Let’s discuss governance. This is critical. I don’t want to fund something that goes in what turns out to be the wrong direction.”

*Exhibit No. 12

In a November 2015 email exchange between Musk and Altman, the two discuss plans for the forthcoming AI lab. Musk starts off by recounting a “great call with Greg [Brockman]” and saying he’s “super impressed with everyone so far,” calling it a “great team.” He suggests creating the lab as an “independent, pure play 501c3, but with a crystal clear focus on the positive advent of strong AI distributed widely to humanity.” He says the company would “still aim to bring in revenue in excess of costs at some point, but positive net revenue would just flow to cash reserves.”

With regard to compensation for employees, Musk suggests a cash salary and certain bonuses. He says that if Altman is amenable, employees could convert cash to stock in Y Combinator, adding that it’s fine if they’d rather convert some or all to SpaceX stock instead. (“I can pretty much do what I want on the SpaceX side, as it is private (thank goodness),” Musk writes.) He also offers “insane amounts of real world sensor data” from Tesla for the AI lab to use, mentioning that the amount of data is “several orders of magnitude greater than any other company.”

Musk’s first stab at a name for the AI lab is “Freemind,” as he says it “conveys the sense that we are trying to create digital intelligence that will be freely available to all — the opposite of Deepmind’s one-ring-to-rule-them-all approach.” He also says he’ll dedicate whatever amount of his time is useful, even though that could mean less time allocated to SpaceX and Tesla. “If I really believe that this is potentially the biggest near-term existential threat, then action should follow belief,” he writes. He adds later that, despite seemingly trying to be essentially a silent partner, he has to “bite the bullet on admitting real involvement. This will come as a shocker to many, but so be it. Can’t be lukewarm about this.”

Altman suggests the AI lab share a building with Y Combinator and use the incubator’s legal team to help get it started. He also suggests the names “Axon” or something related to famed computer scientist and mathematician Alan Turing.

Musk writes, “Something Turing-related that doesn’t sound too ominous might be good. Want to avoid the Turing Test association though, as that sounds too much like we are replacing humans.”

*Exhibit No. 14

A December 2015 email exchange between Altman and Musk drafts the opening paragraphs of OpenAI’s mission and press release. Musk says the “whole point of this release is to attract top talent.” The two go back and forth on wording, and the final product ends up not straying too much from Musk’s original draft.

Musk writes in his draft that “the outcome of this venture is uncertain and the pay is low compared to what others will offer, but we believe the goal and the structure are right.” Altman writes in his draft that “because we don’t have any financial obligations, we can focus on the maximal positive human impact and disseminating AI technology as broadly as possible.”

Exhibit No. 16

OpenAI’s official articles of incorporation, filed December 8th, 2015. The document states that OpenAI “shall be a nonprofit corporation organized exclusively for charitable purposes” and that its purpose is “to ensure that artificial general intelligence benefits all of humanity, including by conducting and/or funding artificial intelligence research. The corporation may also research and/or otherwise support efforts to safely develop and distribute such technology and its associated benefits, including analyzing the societal impacts of the technology and supporting related educational, economic, and safety policy research and initiatives.”

The document continues, “The resulting technology will benefit the public and the corporation will seek to distribute it for the public benefit when applicable. The corporation is not organized for the private gain of any person.”

Exhibit No. 70

An April 2016 email exchange between Musk and Nvidia CEO Jensen Huang. Musk asks Huang if the OpenAI team can buy an early unit of a supercomputer, making sure to highlight that “OpenAI is unaffiliated with Tesla. It is a non-profit funded by me and a few others with the goal of developing safe AGI (and hopefully not paving the road to hell with good intentions).”

Huang responds that he “will make sure OpenAI gets one of the first ones.”

Exhibit No. 388

A photo of Jensen Huang ostensibly dropping off said computer. Elon Musk stands nearby.

On the wall behind him is a lengthy quote sometimes attributed to US Navy Admiral Hyman G. Rickover, which is echoed in a 2013 blog post by Altman. (The Verge couldn’t immediately confirm the whole quote was said by Rickover; in a US Navy post attributed to the admiral, only part of the quote appears: “Man has a large capacity for effort. But it is so much greater than we think it is, that few ever reach this capacity.”)

*Exhibit No. 152

In an August 2017 email exchange between Musk and Shivon Zilis, Musk’s chief of staff who eventually sat on OpenAI’s board, and with whom Musk would eventually share multiple children. Zilis writes a recap of her meeting with Brockman and Sutskever, laying out seven unanswered questions. She says Brockman and Sutskever are fine with Musk spending less time on the company and having less control, or spending more time and having more control, but not less time and more control. They also hope to raise significantly more than $100 million to start, as they worry the data center they need alone would cost that much. She says Brockman is relatively set on an equal equity split. They also, she writes, worry about Musk’s control over the company. In her notes recapping their concerns, Zilis writes, “Is the requirement for absolute control? They wonder if there is a scenario where there could be some sort of creative overrule position if literally everyone else disagreed on direction.”

The biggest point of tension, Zilis writes, seems to be on Musk’s duration of control over the company, despite his ownership stake. “*The* non-negotiable seems to be an ironclad agreement to not have any one person have absolute control of AGI if it’s created. Satisfying this means a situation where, regardless of what happens to the three of them [Greg, Ilya, and Sam], it’s guaranteed that power over the company is distributed after the 2-3 year initial period … An ironclad 2-3yr minority control agreement, regardless of the fates of Greg / Sam / Ilya.”

Musk responds, “This is very annoying. Please encourage them to go start a company. I’ve had enough.”

*Exhibit No. 153

A September 2017 email to Musk from Jared Birchall, an adviser to Musk and manager of his family office. He attaches a “more user friendly version of the cap table that Ilya and Greg are proposing.”

In it, Musk is reflected as having 51.20 percent equity, with Altman, Sutskever, and Brockman each having 11.01 percent. There’s also reserved equity for employees, and the cap table denotes each initial employee’s name or nickname followed by a proposed amount of equity.

Documents released April 30, 2026

*Exhibit No. 25

A November 2015 email exchange between Musk and Altman, in which Altman references what seems to be one of the first names and structures considered for the AI lab — Y Combinator AI.

Altman writes that the “plan is to have you, me, and Ilya on the Board of Directors for YC AI, which will be a Delaware non-profit,” adding, “We will write into the bylaws that any technology that potentially compromises the safety of humanity has to get consent of the Board to be released, and we will reference this in the researchers’ employment contracts.”

Musk disagrees in his response: “I think this should be independent from (but supported by) YC, not what sounds like a subsidiary. Also, the structure doesn’t seem optimal. In particular, the YC stock along with a salary from the nonprofit muddies the alignment of incentives. Probably better to have a standard C corp with a parallel nonprofit.”

Exhibit No. 559

In a December 2016 email exchange between Musk and his Neuralink associates, he brings up his concerns about beating Google Deepmind again, writing, “Deepmind is moving very fast. I am concerned that OpenAI is not on a path to catch up. Setting it up as a non-profit might, in hindsight, have been the wrong move. Sense of urgency is not as high.”

Exhibit No. 773

In June 2017, Musk writes an email saying he hired Andrej Karpathy away from OpenAI to be director of Tesla Vision, saying, “The OpenAI guys are gonna want to kill me, but it had to be done…”

*Exhibit No. 631

In July 2017, Musk writes in an email to Sutskever and Brockman that China “will do whatever it takes to obtain what we develop. Maybe another reason to change course.” Brockman says he agrees, and that the path ahead should be an “AI research non-profit (through end of 2017), AI research and hardware for-profit (starting 2018), [and] government project (when: ??).”

As a token of appreciation for their work at OpenAI, Musk offers to give Sutskever, Brockman, and others on the team Tesla Model 3 cars that are “not available to the public.”

Exhibit No. 642

Musk asks in August 2017 if Altman, Sutskever, and Brockman can meet to discuss the “next step” for OpenAI — and volunteers “the haunted mansion [he] just bought near SF,” although it’s “kinda crazy and weird and will have party carnage.”

*Exhibit No. 662

An email exchange between Musk and Birchall, his money manager, later in August 2017. Birchall writes that for now, he’s “held off” on giving OpenAI Musk’s typical quarterly $5 million donation and asks if he should continue holding off. Musk responds affirmatively.

Exhibit No. 686

A September 2017 email exchange between Musk, Brockman, and Sutskever, with Sutskever suggesting that Musk have three board seats and Brockman, Sutskever, and Altman each have one. Musk responds that he believes he should have the right to appoint four board seats and later compliments the three others.

Musk writes, “I would not expect to appoint [the four board seats] immediately, but, like I said I would unequivocally have initial control of the company, but this will change quickly. The rough target would be to get to a 12 person board (probably more like 16 if this board really ends up deciding the fate of the world) where each board member has a deep understanding of technology, at least a basic understanding of AI and strong & sensible morals.”

Exhibit No. 679

A September 2017 email exchange between Brockman and Musk, with Altman and Sutskever CC’d. Brockman and Sutskever propose a cap table for Musk’s approval, with Brockman noting that himself and Altman are able to invest a lot more than Sutskever, but Sutskever can invest more than $2.5 million if he takes a loan from Altman and/or Brockman securitized by stock he owns.

Musk replies, “Guys, you are pushing too hard here. I’m not ok with this.”

*Exhibit No. 691

A September 2017 text message from Musk to Zilis and others. Musk writes, “We should get going on creating the OpenAI B Corp, as I promised Greg and Ilya. Let’s discuss this eve. Still no word from Sam Altman btw.”

*Exhibit No. 158

A September 2017 email exchange between Altman, Musk, Zilis, Brockman, Sutskever, and Musk’s chief of staff Sam Teller. It paints a picture of a two-sided negotiation with peak tension, with Musk and Altman essentially on one side and Brockman and Sutskever on the other.

To Elon, Brockman and Sutskever write, “Elon: We really want to work with you … Our desire to work with you is so great that we are happy to give up on the equity, personal control, make ourselves easily firable — whatever it takes to work with you.” However, they write they were concerned about Musk’s control over the future technology OpenAI may put out.

“The current structure provides you with a path where you end up with unilateral absolute control over the AGI,” the two write to Musk. “You stated that you don’t want to control the final AGI, but during this negotiation, you’ve shown to us that absolute control is extremely important to you. As an example, you said that you needed to be CEO of the new company so that everyone will know that you are the one who is in charge, even though you also stated that you hate being CEO and would much rather not be CEO. Thus, we are concerned that as the company makes genuine progress towards AGI, you will choose to retain your absolute control of the company despite current intent to the contrary. We disagree with your statement that our ability to leave is our greatest power, because once the company is actually on track to AGI, the company will be much more important than any individual.”

The two also touch on the team’s often-mentioned fears about Deepmind’s Demis Hassabis. To Musk, they write, “The goal of OpenAl is to make the future good and to avoid an AGI dictatorship. You are concerned that Demis could create an AGI dictatorship. So do we. So it is a bad idea to create a structure where you could become a dictator if you chose to, especially given that we can create some other structure that avoids this possibility.”

Brockman and Sutskever have different concerns for Altman himself, though.

In the part of the message directed at Altman, they write, “We haven’t been able to fully trust your judgements throughout this process, because we don’t understand your cost function. We don’t understand why the CEO title is so important to you. Your stated reasons have changed, and it’s hard to really understand what’s driving it.” Separately, they question some of Altman’s motivations, asking him, “Is AGI truly your primary motivation? How does it connect to your political goals? How has your thought process changed over time?”

Altman responded to the email that he “remain[ed] enthusiastic about the non-profit structure!”

*Exhibit No. 157

A September 2017 response from Musk to the above concerns detailed by Brockman and Sutskever. Musk writes, “Guys, I’ve had enough. This is the final straw. Either go do something on your own or continue with OpenAl as a nonprofit. I will no longer fund OpenAl until you have made a firm commitment to stay or I’m just being a fool who is essentially providing free funding for you to create a startup. Discussions are over.”

*Exhibit No. 159

A September 2017 email exchange between Zilis and Musk. Zilis recounts some of Altman’s feelings, like the idea that Altman “lost a lot of trust” for Brockman and Sutskever during the negotiations, feeling that their messaging was “inconsistent” and “childish at times.” She also says Altman was planning to take a 10-day hiatus from OpenAI to think about how much he trusted Brockman and Sutskever and how much he wanted to work with them.

She also says Altman mentioned that Holden Karnofsky — a prominent tech executive and leader in effective altruism, who now works at Anthropic and is married to Anthropic co-founder Daniela Amodei — was “irked by the move to for-profit and potentially offered [a] more substantial amount of money if OpenAI stayed a non-profit.”

Zilis also says that Altman is “great with keeping the non-profit” and that though Brockman and Sutskever are also amenable to continuing with the non-profit structure, “they know they would need to provide a guarantee that they won’t go off doing something else to make it work.”

Exhibit No. 719

An October 2017 email from Musk to his Neuralink co-founder Ben Rapoport. Musk writes, “Hire independently or directly from OpenAI. I have no problem if you pitch people at Open Al to work at Neuralink.”

Exhibit No. 98

On New Year’s Day in 2018, Sutskever writes a note of gratitude to Musk, cc’ing Brockman, calling Musk the “most overwhelmingly competent person in the world” and adding that he’s thankful Musk pushed OpenAI to build custom hardware.

Exhibit No. 99

Brockman sends a similar message as Sutskever did to Musk on New Year’s Day 2018, writing that “it’s an honor to work alongside you.”

*Exhibit No. 749

In a January 2018 email exchange between Musk, Altman, Brockman, and Sutskever, with Zilis CC’ed, Musk writes of his concerns about Google Deepmind’s advancement in AI. He writes, “OpenAl is on a path of certain failure relative to Google. There obviously needs to be immediate and dramatic action or everyone except for Google will be consigned to irrelevance. I have considered the ICO approach and will not support it. In my opinion, that would simply result in a massive loss of credibility for OpenAl and everyone associated with the ICO. If something seems too good to be true, it is. This was, in my opinion, an unwise diversion.”

Musk continues, “The only paths I can think of are a major expansion of OpenAl and a major expansion of Tesla Al. Perhaps both simultaneously. The former would require a major increase in funds donated and highly credible people joining our board. The current board situation is very weak … To be clear, I have a lot of respect for your abilities and accomplishments, but I am not happy with how things have been managed. That is why I have had trouble engaging with OpenAl in recent months. Either we fix things and my engagement increases a lot or we don’t and I will drop to near zero and publicly reduce my association. I will not be in a situation where the perception of my influence and time doesn’t match the reality.”

When Musk forwards the back-and-forth to Andrej Karpathy, Karpathy responds in support of Musk’s thoughts, writing, “Working at the cutting edge of AI is unfortunately expensive … It seems to me that OpenAl today is burning cash and that the funding model cannot reach the scale to seriously compete with Google (an 800B company). If you can’t seriously compete but continue to do research in open, you might in fact be making things worse and helping them out ‘for free,’ because any advances are fairly easy for them to copy and immediately incorporate, at scale.”

Karpathy continues, “A for-profit pivot might create a more sustainable revenue stream over time and would, with the current team, likely bring in a lot of investment. However, building out a product from scratch would steal focus from Al research, it would take a long time and it’s unclear if a company could ‘catch up’ to Google scale, and the investors might exert too much pressure in the wrong directions.”

Karpathy says the “most promising option” he can think of “would be for OpenAl to attach to Tesla as its cash cow. I believe attachments to other large suspects (e.g. Apple? Amazon?) would fail due to an incompatible company DNA.”

He then goes on to detail what a Tesla-OpenAI merge would look like. “Using a rocket analogy, Tesla already built the ‘first stage’ of the rocket with the whole supply chain of Model 3 and its onboard computer and a persistent internet connection. The ‘second stage’ would be a full self driving solution based on large-scale neural network training, which OpenAl expertise could significantly help accelerate. With a functioning full self-driving solution in ~2-3 years we could sell a lot of cars/trucks. If we do this really well, the transportation industry is large enough that we could increase Tesla’s market cap to high O(~100K), and use that revenue to fund the Al work at the appropriate scale. I cannot see anything else that has the potential to reach sustainable Google-scale capital within a decade.”

Musk forwards the note to Sutskever and Brockman, writing that Karpathy is right, and that “Tesla is the only path that could even hope to hold a candle to Google. Even then, the probability of being a counterweight to Google is small. It just isn’t zero.”

*Exhibit No. 761

A February 2018 text message conversation between Musk and Zilis, potentially just after Musk told Altman, Brockman, and Sutskever on a video meeting that he would be departing OpenAI’s board.

Zilis writes, “Do you prefer I stay close and friendly to OpenAl to keep info flowing or begin to disassociate? Trust game is about to get tricky so any guidance for how to do right by you is appreciated.” Musk responded, “Close and friendly, but we are going to actively try to move three or four people from OpenAl to Tesla. More than that will join over time, but we won’t actively recruit them.”

The two discuss who on the team to potentially recruit, with Zilis saying that Sutskever was “visibly devastated” after Musk left the video meeting and that there is “some probability you could get Ilya if you wanted him, but don’t know if you do. He has been a very good spiritual leader.” Musk responds, “There is little chance of OpenAI being a serious force if I focus on Tesla AI.”

Zilis goes on to touch on the often-brought-up fear of Google’s progress in the AI race and tries to encourage Musk to “slow down” Hassabis, CEO of Google Deepmind. She writes, “There is a very low probability of a good future if someone doesn’t slow Demis down. Slowing him down is the only nonnegotiable net good action I can see. You don’t realize how much you have an ability to influence him directly or otherwise slow him down. I think you know I’m not a malicious person but in this case it feels fundamentally irresponsible to not find a way to slow or alter his path.” Musk responds, “I doubt I could do so in a meaningful way,” and says they can speak by phone about it later that evening.

Exhibit No. 233

An April 2018 email exchange between Musk and Zilis, with Zilis writing that OpenAI’s first funding round will likely be “largely Reid [Hoffman, LinkedIn co-founder] money, potentially some corporates.” Zilis also writes that Quora CEO Adam D’Angelo is primed to take Musk’s place on OpenAI’s board. (D’Angelo would later be involved in Altman’s 2023 ouster from his CEO role.)

*Exhibit No. 819

In a July 2018 email to Musk, Zilis updates him on the new funding round OpenAI is planning, as well as a public letter detailing concerns about autonomous weapons that the Future of Life Institute is planning to publish soon, which Musk had been listed as a signatory on in the past.

Zilis also recounts rumors she’s heard about Google Deepmind’s Hassabis, writing, “Rumor has it that, on top of the folks that secretly converse on Twitter DM because they don’t trust Demis not to spy on their email and gchat, a part of the inner group also meets in a London coffee shop without cell phones to have in person discussions away from him. Heard this from both Altman and another friend.”

*Exhibit No. 236

An August 2018 email from Altman to Musk, in which he includes OpenAI’s official term sheet. Altman writes that his “current thought” is that he won’t take any equity in OpenAI. He goes on to say, “I’m not doing this for the money anyway, and I like the idea of being completely unconflicted and just focused on the best outcome for the world. If it appeared at some point we weren’t going to build AGI but were going to build something valuable, then maybe I’d want equity then.”

The term sheet includes a large purple warning box at the top, stating within asterisks, “Investing in OpenAI LP (the Partnership) is a high-risk investment. Investors could lose their capital contribution and not see any return. It would be wise to view any investment in OpenAI LP in the spirit of a donation, with the understanding that it may be difficult to know what role money will play in a post-AGI world.” The term sheet goes on to summarize planned revenue and how technology may be commercialized in the future, as well as the company’s fiduciary duties and planned fundraising.

“Our duty to these principles and the advancement of our mission takes precedence over any obligation to generate a profit,” the term sheet states. “We may never make a profit, and we are under no obligation to do so. We are free to re-invest any or all of our cash flow into research and development activities and/or related expenses without any obligation to the Limited Partners … The fiduciary duties of the Nonprofit Board of Directors flow exclusively to the Nonprofit, not to the Limited Partners.”

*Exhibit No. 844

In November 2018, Musk writes in an email to Gabe Newell, co-founder of video game developer Valve, that his involvement in OpenAI is “very limited at this point.”

“I still provide some financial support and get verbal and email updates every few weeks from Sam Altman, but don’t spend time there,” Musk says. “I lost confidence that OpenAl could muster the resources to serve as an effective counterweight to Google/Deepmind and decided to attempt that through Tesla instead. We have cash flow on the order of billions of dollars per year to build hardware that hopefully has at least a dark horse chance to keep Google honest. Probably worth talking about at some point.”

Newell responds that he’s happy to talk about Tesla and AI when Musk is ready.

*Exhibit No. 853

A December 2018 email exchange between Musk and Altman, with others CC’ed. Musk writes of his intensifying fears about Google Deepmind’s Hassabis taking over in the AI race. “My probability assessment of OpenAl being relevant to DeepMind/Google without a dramatic change in execution and resources is 0%. Not 1%. I wish it were otherwise. Even raising several hundred million won’t be enough. This needs billions per year immediately or forget it. Unfortunately, humanity’s future is in the hands of Demis … And they are doing a lot more than this.”

Musk continues, “OpenAl reminds me of Bezos and Blue Origin. They are hopelessly behind SpaceX and getting worse, but the ego of Bezos has him insanely thinking that they are not! I really hope I’m wrong.”

Altman responds to ask if the two can meet to discuss increasing that percentage. He says he believes OpenAI has a good plan and a good path to gain the capital they need but that they aren’t executing quickly enough. “None of us want to be Bezos here!” he says.

Musk writes, “OpenAl is not a serious counterweight to DeepMind/Google and will only get further behind. It is surprising that this … isn’t obvious to you. In general, always overestimate competitors. You are doing the opposite.”

The two agree to meet in Puerto Rico later that week.

Exhibit No. 239

A March 2019 email exchange between Altman and Musk, with Zilis and Teller CC’ed. Altman sends a blog post detailing OpenAI’s new capped-profit structure to Musk for approval.

Exhibit No. 862

Zilis circles back on Altman’s note above in March 2019, highlighting the part where it says Musk left the board of OpenAI’s nonprofit in February 2018 and that he is not involved with OpenAI LP.

Exhibit No. 863

Altman texts Musk a couple of days later in March 2019, reminding him they’re planning to announce OpenAI’s new structure tomorrow and wanting to check the wording about Musk’s past involvement.

“Also have some mild Demis updates to share,” Altman writes. Musk agrees to talk over the phone soon.

Exhibit No. 869

In April 2019, Altman texts Musk to ask if he has time to talk about Microsoft’s investment in OpenAI.

Exhibit No. PX251

In September 2020, Musk publicly responds to a social media post linking to a VentureBeat article about Microsoft getting the exclusive license to OpenAI’s GPT-3, writing, “This does seem like the opposite of open. OpenAI is essentially captured by Microsoft.”

*Exhibit No. 105

An October 2020 test message exchange between Musk and Altman, with Altman reaching out to say he saw Musk’s posts on social media the prior week about Microsoft’s exclusive license to OpenAI’s GPT-3. Altman writes, “I think there’s no way we can hold a candle to DeepMind without many billions of dollars, and MSFT still seems like the best way for us to get that with the least compromise. We gave MSFT a copy of GPT-3 to use in their own products, but we still get to retain autonomy to release our work ourselves (e.g., we can and will continue to provide API access to the most powerful language model in existence to everyone).”

Musk responds, “Yeah, we should talk. I don’t think it’s a winning approach to be (or at least appear to be) hypocritical. At least change the name.”

Musk later links to a social media post saying that one of Musk’s “worst management blunders” was exclusively licensing GPT-3 to Microsoft. Altman responds saying that OpenAI “finally just got a full time PR person,” name-dropping Apple’s former PR person Steve Dowling as the new hire, and writing, “I am hopeful we can start getting pr right…” Dowling would later step down from his role, which reported directly to Altman, at the beginning of 2021.

Exhibit No. 252

In a text message exchange between Musk and Altman in late October 2020, Altman asks for advice on the next Microsoft investment that OpenAI is considering. Musk responds that he can talk in the next day or two.

Exhibit No. 295

An October 2022 article from The Information about OpenAI’s advanced talks with Microsoft for additional funding.

*Exhibit No. 296

In October 2022, Musk writes in a text message to Altman that he was “disturbed to see OpenAI with a $20B valuation … I provided almost all the seed, A and most of B round funding.” He sends a link to the above article, adding, “This is a bait and switch.”

Altman responds, “I agree this feels bad—we offered you equity when we established the cap profit, which you didn’t want at the time but we are still very happy to do any time you’d like. We saw no alternative, given the amount of capital we needed and needing still to preserve away to ‘give the AGI to humanity’, other than the capped profit structure. Fwiw I personally have no equity and never have. Am trying to navigate tricky tightrope the best I can.” The two agree to talk sometime in the coming week.

Exhibit No. 1025

In March 2023, Musk posts on social media, “I’m still confused as to how a non-profit to which I donated ~ $100M somehow became a $30B market cap for-profit. If this is legal, why doesn’t everyone do it?”

*Exhibit No. 355

A May 2023 text message exchange between Musk, Altman, Birchall, and Musk lawyer Alex Spiro, in which it’s detailed that Spiro, and potentially Birchall, will show up to OpenAI’s headquarters to review documents about OpenAI’s structure and its relationship with Microsoft.

Musk writes, “The point is to understand the relationship between all the companies and the original OpenAI 501c3 … Understanding what rights Microsoft has is important. One of the things I’m concerned about is that they will have de facto control over AGI.”

*Exhibit No. 1444

A March 2026 social media post by Musk. He writes, “Tesla will be one of the companies to make AGI and probably the first to make it in humanoid/atom-shaping form.”

Exhibit No. 1293
A list of “undisputed facts” in Musk v. Altman, et al., including details on timeline and amounts of money raised and/or donated.

Documents released May 1, 2026

Exhibit No. 1284

An agreement establishing a philanthropic account called Musk Charitable at Vanguard Charitable, signed by Elon Musk in July of 2014.

*Exhibit No. 502

An email from Sam Altman to Elon Musk with a list of suggestions for OpenAI, including a governance structure of five people, including Musk, Altman, Bill Gates, Pierre Omidyar, and Dustin Moskovitz. “Agree on all,” Musk responds.

Exhibit No. 17

A December 11th, 2015 blog post titled “Introducing OpenAI” — also available publicly online. The post describes OpenAI as a “non-profit artificial intelligence research company” whose goal is to reach “advanced digital intelligence in the way that is most likely to benefit humanity as a whole, unconstrained by the need to generate financial return.” It lists the founding team, including Sutskever, Brockman, and Andrej Karpathy (who would later go to Tesla), as well as the co-chairs, Altman and Musk.

*Exhibit No. 516

A January 2016 email chain. Musk forwards Sutskever and Altman a message from Google’s Hassabis, where Hassabis objects to Musk, Altman, and others “extolling the virtues of open sourcing AI … I presume you realise that this is not some sort of panacea that will somehow magically solve the AI problem?” Hassabis describes the approach as “actually very dangerous” and links to a Slate Star Codex blog post.

Sutskever responds by saying that “as we get closer to building AI, it will make sense to start being less open” and “totally OK to not share the science (even though sharing everything is definitely the right strategy in the short and possibly medium term for recruitment purposes).” Musk replies: “Yup.”

*Exhibit No. 71

A series of emails involving Musk, Altman, former OpenAI COO Chris Clark, and Ronald Gong (an associate of Musk who’s listed on financial documents). The chain starts in February of 2016, with Altman emailing Musk that “I think we’re going to need more than I was originally budgeting given a) the salaries in the field and b) the speed at which you want to grow.” Musk agrees to contribute $20 million a year for the next three years, while Altman contributes $10 million a year, and $5 million a year comes from other donors. Gong and Clark discuss using YC Org as a fiscal sponsor for OpenAI, and Clark attaches the organization’s articles of incorporation and other documentation.

Exhibit No. 60

Documentation for a series of grants from the Musk Charitable Fund to OpenAI, including a mid-2016 grant of $5 million to YC Org, directed toward the “OpenAI Artificial Intelligence Research Program”; a $4.5 million grant in August of 2016; and a series of monthly $175,000 lease payments in 2017, among others.

Exhibit No. 73

A letter from Chris Clark (listed as the treasurer of YC Org) acknowledging a $500,000 donation from Musk in May of 2016.

*Exhibit No. 532

A May 2016 exchange between Brockman and Musk. Brockman writes that “Google’s policy people want to speak with me,” apparently because they’re afraid they’ll “build a public narrative that it’s wrong to have any closed-source AI.” Brockman says he plans to say there’s no reason to do that. ”We don’t have a problem with people keeping things proprietary — it’s fine to make money off this stuff, and we may even generate revenue ourselves one day,” he says. “What, that’s really interesting. Who called from Google?” Musk asks.

*Exhibit No. 539

A June 2016 exchange between Elon Musk and his fixer, Jared Birchall, discussing a lease of the Pioneer Building in San Francisco (which housed OpenAI until 2024 and xAI after that). Birchall mentions that a lease has been finalized and is awaiting Sam Altman’s signature, and Musk objects: “Since I’m personally on the hook, this should be viewed as a Musk Foundation building, in which we will house OpenAI, Neuralink, and maybe some SpaceX or Tesla people. I don’t want Sam on the lease.” Birchall says he’ll direct Altman’s name to be removed from the lease.

Exhibit No. 75

A June 2016 email chain involving Jared Birchall and two associates of Bridgeton Holdings, Atit Jariwala, and Bourke Lee. The messages negotiate leasing the Pioneer Building and end with instructions for making the first monthly lease payment of around $142,000.

Exhibit No. 545

A June 2016 email exchange between Altman, Birchall, and Clark about financing the Pioneer Building lease. Clark sends Birchall a Tenancy at Will agreement signed by Altman, attached to the email.

Exhibit No. 79

A July 1st email from Birchall to Musk with the executed lease to the Pioneer Building, including the lease. Birchall notes that the building owner will “facilitate a site inspection as soon as we’d like.”

*Exhibit No. 80

A July 2016 email chain between Musk and Birchall. Birchall sends Musk details about the quarterly donations and monthly rent payments for OpenAI, plus a request from Clark, who “asked me about using the extra space in the building for some of the Y Combinator companies.” Musk’s response mentions that “I have had very little bandwidth to think about the company and am a little worried that it is being managed as an extension of Y Combinator” and says he’d also like to use part of the building for Neuralink, “so no YC stuff.”

Birchall then says there was a problem with the first quarterly contribution: “because they didn’t have an entity in place to even make a contribution we didn’t pay,” and in June they began using another nonprofit (presumably YC Org) as a conduit. “I’m not sure why they have taken so long to apply,” Birchall complains. “So I haven’t sent anything to OpenAI? That’s a really big deal. My credibility is at stake here,” Musk writes. Birchall confirms the funds were sent — just channeled through a temporary 501(c)(3). “Good,” Musk answers.

*Exhibit No. 556

An August 2016 email exchange between Musk and Altman. Altman tells Musk he’s negotiated a $50 million compute donation from OpenAI over the next 3 years and asks if there’s any reason to care about switching from Amazon. “I’m ok with this only if they don’t use it in marketing. I would also like to see the exact terms and conditions. Gifts are only as good as the T&C,” Musk writes. “I think Jeff [Bezos] is a bit of a tool and Satya [Nadella] is not, so I slightly prefer Microsoft, but I hate their marketing dept.”

Altman writes that “Amazon started really dicking us around on the T+C, especially on marketing commits. And their offering wasn’t that good technically anyway.” Musk says that “I will call Satya if we get to decent terms” and says that Microsoft can always point people to “a simple text blog expressing appreciation of Microsoft’s donation on our website.”

Exhibit No. 84

A series of emails between October and November of 2016 involving Birchall; Gong; Morgan Stanley Private Wealth Management group director Matilda Simon-Ferrigno; and two people from Gong’s company myCFO, Teresa Holland and Paula Lo. Birchall arranges moving shares from the Musk Foundation to finance OpenAI.

*Exhibit No. 51

OpenAI’s 2016 tax returns as a 501(c)(3) nonprofit. It lists 52 employees and around $13 million in total revenue, mostly from contributions and grants. It names accomplishments including establishing a research team, launching the OpenAI Gym Beta, publishing “nearly half a dozen comprehensive research papers,” holding a conference, and building a safety team.

Exhibit No. 810

The Musk Foundation’s 2016 Return of Private Foundation tax documents, showing a total of around $47.8 million in contributions, gifts, and grants.

Exhibit No. 86

A March 2017 letter from Chris Clark to Elon Musk, acknowledging a gift of $5 million to OpenAI via YC Org.

Exhibit No. 87

A June 2017 letter from Chris Clark to Elon Musk, acknowledging a gift of $5 million to OpenAI via YC Org.

Exhibit No. 621

A June 2017 Fidelity charitable investment advisor program application for the Musk Foundation Charitable Fund.

Exhibit No. 92

Emails between Birchall, Clark, and UBS wealth management associate Leeder Hsu in July of 2017. Birchall directs a grant of $250,000 to YC Org for a Universal Basic Income study.

*Exhibit No. 93

A July 2017 email chain involving Brockman, Musk, Sutskever, and Birchall. Musk sends a link to a New York Times story about Chinese AI with the comment, “They will do whatever it takes to obtain what we develop. Maybe another reason to change course.” Brockman suggests a path of an AI research nonprofit through 2017, “AI research + hardware for-profit” starting 2018, and “Government project (when: ??).” Musk then says that “in appreciation for what you’ve done to get OpenAI to where it is today,” he’d like to offer some OpenAI founding members Tesla Founder Series Model 3 cars. Birchall says he’ll reach out with details about the cars.

Exhibit No. 646

An August 2017 email conversation between Zilis and Birchall about filing for a for-profit branch of OpenAI. “Elon wants to have control to prevent this from going squirrely,” Zilis says. She lists “unknowns,” including leadership of the new entity — ”Greg 100% doesn’t want to run it.” Birchall sends confirmation of how much Musk gave to OpenAI in 2016 and 2017: $15.4 million and $16 million, respectively.

Exhibit No. 693

OpenAI, Inc.’s certificate of incorporation on September 15th, 2017, as a public benefit corporation.

Exhibit No. 52

OpenAI’s 2017 tax returns, also as a 501(c)(3) nonprofit. It lists around $33 million in revenue (mostly from contributions and grants, again) and 99 employees. It notes that in 2017, it demonstrated that “reinforcement learning algorithms could be scaled to beat the world’s best humans at a restricted version of an advanced, multiplayer game called Dota 2.”

Exhibit No. 1285

A copy of the Vanguard Charitable Policies and Guidelines, 2014 to 2017.

Exhibit No. 91

Documentation for a series of 2017-2020 donations from Musk to OpenAI, composed of monthly “general support” payments that likely include the Pioneer Building lease — which Musk said constituted his main form of support in the later years of OpenAI.

*Exhibit No. 100

A January 2018 letter from Clark to Musk acknowledging a gift of four Tesla sedans with a total value of around $250,000.

Exhibit No. 24

The OpenAI Charter from April 9th, 2018. It outlines “the principles we use to execute on OpenAI’s mission,” including “broadly distributed benefits,” “long-term safety,” and “technical leadership.”

*Exhibit No. 827

An August 31st, 2018 email from Altman to Musk with a for-profit Limited Partnership term sheet attached. “Please see attached, look forward to feedback,” Altman says. He says that “my current thought is that I won’t take any equity,” since he likes the idea of “being completely unconflicted,” but says that if OpenAI appeared unlikely to build AGI but “were going to build something valuable, then maybe I’d want equity then.” At the start of the term sheet is a box marked “Important warning,” saying that the partnership is a “high-risk investment” and any investment should be “in the spirit of a donation.”

*Exhibit No. 828

An August 31st, 2018 email from Zilis to Birchall, forwarding Altman’s email. Birchall responds: “Pretty plain vanilla for-profit structure. So kinda hard to push a narrative that doesn’t involve investors being very focused on ROI. I’m a super fan of capitalism and making tons of money doing great things, but not sure if this correlates with the whole ‘noble cause for humanity, not doing it to make money’ narrative.”

Exhibit No. 103

A July 2020 email from Clark to Birchall confirming that OpenAI’s for-profit entity will take over rent payments and suggesting a final one-time donation for security costs and “anticipated landlord project passthrough” of $570,000. “We certainly understand if you’d prefer to just stop everything now,” Clark says, telling Birchall to “do whatever you feel is most fair.”

*Exhibit No. 849

A November 2018 text message chain between Birchall and Greg Smithies, then Neuralink and the Boring Company’s finance head. It discusses a disagreement over rent payments between OpenAI and Neuralink — Smithies says “I’d expect [OpenAI] to get pretty nasty about it (ie probably willing to sue) if we didn’t pay something that they could point their auditors to,” saying “the main driver” is OpenAI accountants demanding it “so they can pass non-profit audits.” Birchall says he’ll “touch base with Chris to get his perspective.”

Exhibit No. 857

A January 2019 message chain between Musk and Birchall, concerning a reimbursement request from OpenAI for shared expenses with Neuralink in the Pioneer Building. Musk offers $250,000 and $1 million in payments for 2017 and 2018, respectively.

Exhibit No. 112

A full list of Elon Musk’s contributions to OpenAI, with entries dating from May of 2016 to September of 2020.

Exhibit No. 1256

A copy of the Fidelity Charitable Policy Guidelines, 2017 to 2022.

*Exhibit No. 1083

An iMessage conversation from December 2024 between Musk and Meta CEO Mark Zuckerberg. Zuckerberg offers a “quick heads up that Meta sent a letter to the California AG supporting your lawsuit against OpenAI. Someone (not us) leaked leaked the letter and it will be public in the next hour. Wanted to make sure you heard this from me.” Musk replies: “Ok.”

*Exhibit No. 1156

A February 2025 iMessage conversation between Musk and Zuckerberg. “Are you open to the idea of bidding on the OpenAI IP with me and some others?” Musk asks. Zuckerberg asks to discuss live, and Musk says, “Will call in the morning.”

*Exhibit No. 1157

A letter from Musk’s xAI and several other investors to OpenAI, proposing an acquisition of all OpenAI’s assets.

Documents released May 2, 2026

Exhibit No. 1092

A late November 2017 email conversation between Zilis, Clark, and Birchall. It includes Clark’s quotes for a security staff and building access controls. “Were any of these for armed coverage?” Birchall asks. Clark says that “I don’t think” it does, but asks if Birchall would prefer that. “Yes, as the profile of the companies grow and as E spends more time on site, we want to make sure we are erring on the side of caution,” Birchall replies.

Documents released May 4, 2026

Exhibit No. 8

In October 2015, Brockman reaches out to potential funders for the new AI nonprofit, name-dropping current investors like Musk, Peter Thiel, Jessica Livingston, and Reid Hoffman. In an email to a “Marissa,” he says that any amount is appreciated, as he himself plans to donate $100,000 and Altman will be donating $100 million. He sends an email BCC’ing Altman, seeming to draft out answers to potential (or maybe already-asked) investor questions. One of these questions is who the chief scientist would be. “We have an extremely well-respected person lined up as chief scientist,” Brockman writes, apparently referencing Sutskever. “He’s currently extricating himself from his employer.”

Exhibit No. 508

A chain of texts between Brockman and Sutskever on November 21st, 2015. “Elon might spend half a day a week with us,” Sutskever says. “I imagined how it will be and I worry that our work environment can become very stressful,” and “since he’ll be bankrolling it it’ll be hard to stop it.” Brockman responds: “I talked about this explicitly with Sam (who I trust on this), I think it’ll be fine. Talking to Elon tomorrow, will let you know how it goes.” Sutskever asks Brockman to “be sure to remind him that research takes longer than one might expect, and that deepmind is a relaxed environment.”

*Exhibit No. 509

A series of November 2015 emails between Musk and Brockman discussing compensation and recruitment for staff at their new AI venture, as well as potential names. Brockman says he and Altman settled on a top three of “Axon,” “AI Summer,” and “Difference Engine.” Musk says “Axon” (which would incidentally soon become the rebrand of Taser) sounds too much like the oil company Exxon, and suggests the name “Freemind.” “Kinda like ‘Freeman’ too, as it reminds me of the scientist protagonist in Half-Life, who was an awesome character, and it sounds like what we are essentially trying to achieve, which is maximum freedom of action for humanity,” he writes.

Brockman also expresses concern about being outfunded by Google, and Musk replies that while “the operating standard should be one of frugality,” even a billion dollars of funding would leave them “massively outmanned and outgunned by Google and Facebook, but we will have right on our side and that counts for a lot.”

*Exhibit No. 66

In November 2015, Brockman attempts to recruit Diederik P. Kingma, known as “Durk” (who, as of time of writing, currently works at Anthropic). Kingma initially says he’ll be going with a different position, but Brockman asks if he’ll chat with Musk before considering his decision final.

Kingma says his main concerns are about ethics, and he adds that as long as he works at the right place, he doesn’t consider it unethical to contribute to AI research. “AI technology will eventually have huge positive or negative impact on society,” he writes. “Therefore, the ethical aspects should not be neglected when making career choices … One big concern is the possibility of an apocalyptic AI scenario, and being partially responsible for it.”

The two go back and forth about the ethics of where and how “human-level AI” might be developed, but Brockman extols the virtues of a nonprofit AI lab.

“On the world’s current trajectory, AGI seems most likely to come out of a for-profit company, or maybe a government intelligence agency as part of an arms race,” Brockman writes. “We have an opportunity to set the stage so that doesn’t happen: I think it should happen as an international coalition, more like the ISS than the space race. The lab’s mission is ensuring that AGI is beneficial, and that those benefits are distributed widely to the world rather than making anyone into a quadrillion-dollar company or omnipotent surveillance state.”

Exhibit No. 67

In December 2015, Musk writes in an email to a group including Sutskever, Brockman, Altman, and Kingma that the OpenAI team is “outmanned and outgunned by a ridiculous margin by organizations you know well, but we have right on our side and that counts for a lot.” He says he’s happy to help with recruitment in any way.

Exhibit No. 521

A late February to early March 2016 email chain involving Musk, Brockman, Sutskever, and SpaceX’s Sam Teller, among others. It describes setting up a meeting between Musk and key OpenAI members every two weeks to get “his lessons and experience” from having “built incredible organizations.”

Musk responds that “I think I should probably do this every week. My dinner with Demis [Hassabis of Google Deepmind] was extremely alarming. I feel like they are playing the Super Bowl and we are playing the Puppy Bowl. Unless we want to have our ass handed to us, we need to step up our game dramatically.” The thread ends with Teller saying they’ll “have to cancel Elon’s visit” due to a SpaceX launch a couple of weeks later.

*Exhibit No. 74

In May 2016, Brockman asks Musk to sign an O-1 visa letter for an OpenAI designer; Musk agrees. But he adds, “I’m having some uneasy feelings about Ilya btw. When I send you an email that doesn’t copy him, that’s intentional. We should talk about this privately.”

Brockman responds, “Sorry to hear that re: Ilya. Happy to chat at any time.”

Exhibit No. 90

In June 2017, Musk tells Brockman, Sutskever, and Altman that he just spoke to Microsoft CEO Satya Nadella about OpenAI needing 10,000 servers with the latest Nvidia GPUs. He says that Nadella said he’d get back to him soon.

“What ideally do we want Microsoft to do?” Musk goes on to say. “Sounds like there is a good chance they will do it.”

*Exhibit No. 350

An August 18th, 2017 email exchange between Musk, Birchall, and Brockman. Birchall emails Musk to describe a “fairly extensive conversation” with Altman, in which “one thing worth mentioning now is that he compensated Greg [Brockman] on the side, giving him a percentage ownership of the LLC that holds the assets of Sam [Altman]’s personal family office – a stake which Sam says equates to ~$10M. Naturally Greg is going to have a greater allegiance toward Sam as a result of this arrangement.” Musk forwards the email to Brockman with one line: “??”

Brockman replies: “There’s no personal loyalty to Sam here — we ran out of OpenAI’s YC stock as we hadn’t tracked the allocation closely enough while making offers, and this was fulfilling my original offer. I feel equal loyalty to both of you.” Brockman continues that “I’m not that motivated by money”, but “I *am* motivated by public recognition for my own work, as I mentioned before the Dota tournament.”

*Exhibit No. 151

Brockman’s journal entry from August 21st, 2017 about wanting to be a billionaire.

He writes, “Ok so what do I *really* want? I want to want to be an engineer. But – think now is a crazy shot to be the one in charge and to step up to the challenge … This is the only chance we have to get out from Elon. Is he the ‘glorious leader’ that I would pick? We truly have a chance to make this happen. Financially what will take me to $1B?”

“Accepting Elon’s terms nukes two things: our ability to choose (though maybe we could overrule him) and the economics. Some chance that rejecting Elon will actually lose us Sam … I’m taking on a lot of pain and suffering. To get 10% of the thing at a billion, we’ re talking $100m.”

Exhibit No. 154

A September 2017 journal entry from Brockman. He outlines his “latest thought process” as being “not happy with the way [Musk]’s steamrolling sam” and worrying that “honestly i just defer to him and say nice things and never push back on anything. i wouldn’t respect me very much in his shoes.”

The file describes ongoing conversations with Musk, Altman, Sutskever, and others over who will maintain control over OpenAI and serve as CEO, as well as the risks of AGI — ”i feel fear. don’t know about you guys,” he recounts Musk saying at one point. It concludes with an email from Musk, who proposes a corporate structure where he would get to a 25 percent influence level on the board, “my min comfort level.” “I’ve been really impressed with the quality of discussion with you guys on the equity and board stuff,” Musk writes. “I have a really good feeling about this.”

*Exhibit No. 161

A November 6th, 2017 journal entry from Brockman. “ilya feeling like we morally should not be kicking elon out, and should be trying to make the non-profit work and convince him to stay,” it begins.

Brockman outlines debates between himself, Altman, and Sutskever about whether to make a deal with Tesla, found a B-Corp, or keep going with a nonprofit. “can’t see us turning this into a for-profit without a very nasty fight,” he writes. “happy to not become rich on this, so long as no one else is,” he says elsewhere, adding “don’t see story for how the tech ends up being owned by the world.” And he notes repeatedly that moving to a for-profit structure would change the “motivations” at OpenAI.

“shit, i feel so good,” begins a final entry, which recounts a meeting where Musk seemed receptive to getting “more results in the non-profit and to fundraise there.” At the end, he says: “btw another realization from this is that it’d be wrong to steal the non-profit from him. to convert to a b-corp without him. that’d be pretty morally bankrupt. and he’s really not an idiot.”

Exhibit No. 163

A November 12th, 2017 journal entry from Brockman. Brockman writes “we’ve been thinking that maybe we should just flip to a for profit. making the money for us sounds great and all.” He adds that “we were a bit too rushed on the for-profit transition before” and “people are being whiplashed by that.” He outlines an OpenAI meeting about pitching for investors, periodically mentioning promoting the fact that OpenAI is a nonprofit as a plus. “so why is this important for people to buy into?” he asks at the end of the entry. “scientific applications, we can *speed it up* and we can ensure it’s not turned by google toward google purposes, owned by the world, public good.”

*Exhibit No. 164

A January 31st, 2018 email exchange between Brockman and Musk. Musk forwards an email from Karpathy, noting that “Google is dominating” paper submissions at a recent deep learning conference. “OpenAI is on a path of certain failure relative to Google,” Musk writes. “There obviously needs to be immediate and dramatic action or everyone except for Google will be consigned to irrelevance.” Musk says he has “considered” an initial coin offering (ICO) but that it would “result in a massive loss of credibility for OpenAI,” and the only options he sees are a massive expansion of OpenAI or a major expansion of Tesla AI. “I have a lot of respect for your abilities and accomplishments, but I am not happy with how things have been managed,” Musk says. “Either we fix things and my engagement increases a lot or we don’t and I will drop to near zero and publicly reduce my association.”

Brockman responds that “I have always been impressed by your focus on the big picture, and agree completely that we must change trajectory and achieve our goals.” He pushes for a major expansion of OpenAI, saying it must build custom AI hardware, a “massive AI data center,” and the “best software team” in the next three years. “Our biggest tool is the moral high ground,” he continues. To retain it, OpenAI must “try our best to remain a n on-profit” and “be perceived as a place that provides public good to the research community.”

Exhibit No. 241

A March 11th, 2019 OpenAI blog post titled “OpenAI LP,” describing the creation of a new hybrid for-profit/nonprofit structure. In a section titled “Who’s involved,” it notes that Musk “left the board of OpenAI Nonprofit in February 2018 and is not formally involved with OpenAI LP.”

Exhibit No. 359

A late December, 2023 email exchange between (now-ex) OpenAI board member Larry Summers and Brockman. Summers suggests issuing a report on “meeting public responsibilities and the like” to meet the requirements of a public benefit corporation, and Brockman sends a draft of a “Charter 2.0,” describing OpenAI’s mission to “provide AGI that benefits all of humanity.”

Exhibit No. 390

A partially redacted April 29th, 2026 court filing detailing Brockman’s supplemental responses and objections to requests for information by Musk. It lays out information about OpenAI’s dealings with companies Brockman holds financial interests in — including CoreWeave, Helion Energy, Stripe, and Cerebras Systems.

Exhibit No. 1201

An undated photograph of Sutskever and other OpenAI staff in a conference room.

Documents released May 5, 2026

Exhibit No. 15

A December 8th, 2015 email sent to Brockman from a redacted party on behalf of Altman, with the subject line “thoughts? from elon.” It’s a brief description of OpenAI’s mission, ending with a final paragraph that reads “The outcome of this venture is uncertain and the pay is low compared to what others will offer, but we believe the goal and the structure is right. We hope this is what matters most to the best in the field.”

Exhibit No. 622

A July 2017 email exchange between Brockman, Altman, Musk, and Sutskever. Brockman writes, “We see a path to building AGI in 5 years, several years before everyone else. This will require acquiring or otherwise merging with Cerebras, and for Cerebras to ship no more than 6-12 months behind schedule. If Google instead acquires Cerebras and Cerebras ships, then success is impossible.”

Brockman goes on to state that the team has “good reason to believe that DeepMind thinks that AGI is ~10 years away. Thus, at least for a while, we have a temporary advantage over DeepMind, Google, and everyone else.”

Musk responds that, according to Jim Keller, an engineer who worked at Tesla and Apple, Cerebras “always oversell[s] their abilities” and that he’d prefer to first see what the Tesla chip team can accomplish.

*Exhibit No. 626

A group of text messages from July 2017 between Zilis and Brockman. Brockman recounts a meeting with someone — potentially Musk — where they discussed OpenAI’s structure. Brockman writes that the person “said non-profit was def the right one early on, may not be the right one now.” Brockman says that he and Sutskever “agree with this for a number of reasons.”

Zilis says the person (likely Musk) is going to Sun Valley to ask Bill Gates to donate; later in the conversation, she says the meeting with Gates went well.

Exhibit No. 632

A July 2017 follow-up to earlier emails previously shared, in which Brockman suggests that an Emma Gallagher from Tesla add Altman to a planned meeting.

*Exhibit No. 653

Two text messages from August 2017, sent by Sutskever to Brockman. Sutskever writes, sarcastically, “At least we’re getting our Teslas! Will a Model 3 make you be willing to accept massively unfavorable terms?”

Exhibit No. 1250.16

A text file from Brockman’s diary dated August 18th, 2017. It contains a single paragraph reading “have been thinking hard about what control really means and whether we should do it. the answer is emerging: definitely not unilateral control. no person should have control over what we’re creating. and what’s fair is an equal split, and we can’t agree to anything less than that.”

Exhibit No. 1250.52

A text file from Brockman’s diary dated November 5th, 2017. It contains bullet points reading “real decision is to fire elon.” and “ok, so we seem converged on the ‘fire elon’ route.”

Exhibit No. 734

A January 2018 email from Brockman to Musk, with Sutskever and Altman CC’ed. Brockman asks Musk to “help close” a list of potential investors he’d listed out, such as Reid Hoffman and Gabe Newell, as well as “entice other[s]” such as Vinod Khosla and Yuri Milner.

*Exhibit No. 200

A limited partnership agreement for OpenAI, dated October 10th, 2018. A prominent box marked “Important” at the top notes that “OpenAI, L.P. is a high-risk investment” (emphasis in original) and that “It would be wise to view any investment in OpenAI, L.P. in the spirit of a donation, with the understanding that it may be difficult to know what role money will play in a post-AGI world.”

*Exhibit No. 748

An early 2018 email exchange between Musk, Brockman, and others. Brockman gives Musk a list of potential board members for OpenAI and writes that over the following three years, OpenAI would need to build custom AI hardware, a “massive” AI data center, and the “best software team.” He also brings up DeepMind, writing that many of DeepMind’s “600+” employees aren’t specifically working on AGI, whereas “every single one of our people will be working on AGI.”

Brockman added, “Our biggest tool is the moral high ground. To retain this, we must: Try our best to remain a non-profit. Al is going to shake up the fabric of society, and our fiduciary duty should be to humanity.” He also noted that OpenAI must “put increasing effort into the safety/control problem … It doesn’t matter who wins if everyone dies. Related to this, we need to communicate a ‘better red than dead’ outlook — we’re trying to build safe AGI, and we’re not willing to destroy the world in a down-to-the-wire race to do so.”

Musk writes to Brockman that he’s “still really concerned that we won’t be able to grow fast enough. Frankly, unless we are able to pull significant numbers of key people from Google/Deepmind, we should assume failure.”

He also goes on to talk about Jeff Bezos and Musk’s view that Blue Origin is behind in the space race, writing, “It feels like we are similar to Blue Origin. Bezos is still clueless as to how hopelessly far behind he is and constantly rationalizes his position, overestimating his ability and dramatically underestimating SpaceX year after year. We lost an excellent engineer to BO last year (first one in several years) and were very concerned. That guy recently returned to SpaceX and told everyone how hopeless they were. Now zero people are even taking BO recruiting calls. Doesn’t matter what they offer.”

*Exhibit No. 202

The agreement between Microsoft and OpenAI that was struck around Microsoft’s $1 billion investment, dated June 28th, 2019 and effective July 2nd of that year. It includes the first version of the companies’ now-defunct AGI clause, which set major components of the deal to end when OpenAI achieved artificial general intelligence. It also defines AGI as “a highly autonomous system that outperforms humans at most economically valuable work.”

Exhibit No. 201

An amended and restated limited partnership agreement for OpenAI, dated July 2nd, 2019. It includes a similar warning to the 2018 agreement, stating investments should be made “in the spirit of a donation.”

Exhibit No. 899

A blog post from OpenAI dated June 11th, 2020. It’s titled “OpenAI API” and describes the release of the company’s API, providing “a general-purpose ‘text in, text out’ interface, allowing users to try it on virtually any English language task.”

*Exhibit No. 204

An “Amended and Restated Joint Development and Collaboration Agreement” between Microsoft and OpenAI, dated March 5th, 2021. The 64-page agreement builds on the one from 2019 and has more specifics about the companies’ agreement for if and when AGI is achieved, including allowing OpenAI to declare some models “sufficient AGI.”

The agreement states, “OpenAl may during the Term, in good faith, declare certain Models developed by LP or OpenAl to be ‘Sufficient AGI’, meaning that such Models: (A) are within the scope of the definition of AGI; and (B) are clearly capable of generating the Target Redemption Amount (defined in the Second Investment Agreement) of return on investment capital for each of OpenAl’s limited partners (including Microsoft).”

If “sufficient AGI” were achieved, OpenAI and Microsoft may collaborate to monetize it until OpenAI generated and paid off the target redemption amounts to each limited partner. After that point, OpenAI and Microsoft were supposed to “collaborate in good faith to make the benefits of artificial general intelligence available to all humanity (‘AGI Distribution’).” Microsoft was also supposed to have the right of first refusal to be OpenAl’s “commercialization partner for AGI” if OpenAI did choose to have one of those. So, throughout the terms of the agreement, OpenAI was to use Microsoft as the “preferred compute partner for AGI,” and Microsoft was not to compete with OpenAI because the contract stated the company would “not pursue AGI (including investments in or collaboration with third parties pursuing AGI).”

Exhibit No. 203

A limited partnership agreement for OpenAI, dated March 6th, 2021. Under the agreement, “Non-Profit” is defined as “OpenAI, Inc., a Delaware nonprofit nonstock corporation.”

Exhibit No. 207

A second “Second Amended and Restated Joint Development and Collaboration Agreement” between OpenAI and Microsoft, dated January 20th, 2023.

*Exhibit No. 206

An amended and restated limited liability company agreement for OpenAI, dated January 23rd, 2023. It has a specific definition, now, for the “Declaration of Sufficient AGI,” meaning “the effective time of the Non-Profit’s determination, in its reasonable discretion, that (x) AGI has been created that has the capability to generate the Target Redemption Amount for each Member; and (y) the Company has the ability and authority to direct the AGI to generate such Target Redemption Amounts. The Company shall promptly notify the Members of such determination.”

In other words, part of the definition of “sufficient AGI” was tied to how much money it may be able to generate.

*Exhibit No. 211

A document concerning the proposed recapitalization of the “OpenAl For-Profit Enterprise,” dated September 11th, 2025.

It notes that the “guiding principle for OpenAl, Inc. (the NFP) is and continues to be the charitable mission, which is to develop artificial general intelligence for the benefit of all humanity,” but that the “expansion of the OpenAl for-profit enterprise and evolution of the AI competitive landscape, among many other factors, have presented unique challenges for the NFP and its ability to further the mission. It is critical to the success of the mission that the for-profit enterprise remain a technical leader in developing AGI, as well as advancing its use by individuals and enterprises. In the view of the NFP Board, the current structure hinders its ability to fulfill the mission, and a change in the structure of the group is essential.” The document proposes a restructuring with a newly formed public benefit corporation, which would be controlled by the nonprofit.

“The mission of the PBC will be identical to the mission of the NFP,” it states. “The transaction also contemplates significant control by the NFP over safety and security matters of the OAI for-profit enterprise, as well as full decision making alignment with the mission with respect to those matters.”

Exhibit No. 212

An undated agreement between OpenAI and Microsoft stipulating further parts of their AGI deal, including the idea that formally declaring AGI would require a panel of five experts — specifically two economics experts, two AI experts, and one legal expert.

It also allows OpenAI to withhold some “AGI Relevant Research” from Microsoft, meaning “confidential, non-public research that, in OAl’s good faith judgment, is reasonably expected to

materially advance the state of the art in a manner that enables a model to satisfy the definition of AGI.”

Exhibit No. 1203

An undated photograph of Brockman taking a group selfie with a crowd in an office.

Exhibit No. 1204

An undated photograph of Brockman taking a selfie in a conference room filled with people.

Exhibit No. 1205

An undated photograph of Brockman and Sutskever posing against a backdrop of several paintings.

Documents released May 6, 2026

Exhibit No. 619

A June 2017 text message conversation between Zilis, Teller, and Gallagher about OpenAI cofounder Andrej Karpathy. “Signed offer in hand for Andrej,” Gallagher writes. “When does E plan to announce to team? Any sense of how mad OpenAI will be? :)” says Teller. Zilis says Karpathy has “talked to a few people” at OpenAI about it. The thread ends a week later, when Teller writes “Autopilot regression. He is going to fire Lattner this morning.”

Exhibit No. 1291

A July 12th, 2017 text conversation between Brockman and Zilis. They discuss a meeting with Musk and a “reverse merger” with a company that is described in testimony as Cerebras. Brockman thanks Zilis for her involvement. “Really useful to have a tighter feedback loop to E and feel like I have a much better perspective and thoughts as a result of our convos,” he says.

Exhibit No. 627

An email from Zilis to Musk and Teller on July 18th, 2017. It describes plans for a “conversation with Greg [Brockman] and Ilya [Sutskever]” that will include confirming OpenAI’s mission as “build AGI before anyone else,” a path that will “involve being much less ‘open’ and require decisions that would prioritize speed over other considerations (safety, government involvement, etc.).” It also mentions plans around hardware, cerebras, and “Dota supremacy… but at what cost.”

Exhibit No. 643

An email from Zilis to Teller and Gallagher dated August 12th, 2017. It includes notes for a “productive” and “contentful” meeting, including “switch to for profit in next couple of weeks (woah, fast!)”.

Exhibit No. 644

An iMessage conversation on August 15th, 2017 between Gallagher, Teller, and Zilis. Zilis describes plans for a “full discussion on immediate next step on conversion to for profit,” as well as the Dota project and Cerebras.

Exhibit No. 651

An email from Zilis apparently to Teller on August 20th, 2017, describing a meeting that includes Brockman, Sutskever, and Altman, apparently including discussion of Musk’s demand for control of the company.

“Greg, Ilya, and Sam have wondered aloud about how you think about [control vs. equity] and if there is any difference. Greg and Ilya seems less sensitive to short-term control than they are long-term control and having a solid amount of equity. Sam seems less sensitive on equity and far more sensitive on control but has been by far the most flexible of the bunch so far.” (Zilis writes, “Unsure if any helpful since it’s mostly soft humany things.”) Teller responds with thanks and suggests telling Musk over the phone.

*Exhibit No. 657

An August 27th, 2017 email from Zilis to Teller and Birchall, describing outstanding questions from a “two hour meeting” with Brockman and Sutskever. “Does Elon require absolute control?” it begins. Further notes state a demand for “an ironclad agreement to not have Elon (or anyone) have absolutely control of AGI they create.” They also “don’t really know how Elon spends his time at the other companies and what he would want to do with OpenAI.”

Exhibit No. 664

An August 30th, 2017 email from Zilis to Teller with a bullet-point list of the founders, governing structure, and operating structure of OpenAI. It lists Musk’s role as “Executive Chairman? CEO?” and Altman’s as “CEO, Chairman, none of the above?” And it describes two options for structure: “Roll everything into a B Corp” and “OpenAI C Corp and OpenAI non-profit.”

Exhibit No. 669

An iMessage conversation from September 1st, 2017 between Teller and Zilis. “Did Elon say to Greg and Ilya that he was stopping OpenAI payments for now?” Teller asks. “Yes,” Zillis says. “He asked when they were going to leave. They said they weren’t planning to, and he said until resolved funding held.” Teller says he’s “concerned about OpenAI sitch” and says they’ll have to be “thoughtful” about setting up calls — ”we should figure out how to best manage. I will think on it,” Zilis says.

*Exhibit No. 707

An email thread from September 26th and 27th, 2017 between Zilis, Musk, and Teller. Zilis says Clark has resigned from OpenAI’s board, and “the removal was always intended (and overdue),” leaving a five-person board that includes Musk, Altman, Brockman, Sutskever, and Open Philanthropy’s Holden Karnofsky. Musk responds that “Board seats need to at least roughly match the proportion of funding provided … I don’t need a majority, but will be ok with half the board vote.” He tells Zilis to relay that “this is a condition of resuming funding.”

Teller suggests he, Zilis, and Birchall join the board to give Musk four of eight board seats. “Yeah for now,” Musk agrees. “My need to give up seats later if we get high profile AI leaders on board.”

*Exhibit No. 708

A September 28th, 2017 email conversation between Teller and Zilis, with the subject “Greg and Ilya.” Zilis describes a meeting where “they took [Musk’s board proposal] in stride” and said the only non-starter included “the transfer to a long-term control structure that included many people, not just Elon. They were scared it wouldn’t happen and needed an ‘iron clad’ structure that couldn’t be gamed.”

The attendees apparently also raised objections to Musk’s stance on for-profit versus non-profit. “it sounds like he expects that all of the things he wanted for the for-profit he also wants for the non-profit, which sucks because the for profit was clearly the better structure,” Zilis quotes them saying. She says they asked if Musk was still open to a for-profit, at which point Zilis “said I had no idea.”

Exhibit No. 710

A September 28th, 2017 email from Zilis to Birchall. “They say they will not move forward without a guarantee to switch away from having control. You and I can argue that’s stupid all we want, but they are holding firm on it,” she says. She suggests a series of possible solutions including a clause that switches control after two years, Musk ending his funding, a breakup, and “Jared’s creative ideas.”

Exhibit No. 712

An October 1st, 2017 email from Zilis to Teller, offering an “honest reflection” of OpenAI executives’ views. It describes Brockman and Sutskever’s “ideal outcomes” as, in order of preference: “Revive the B-Corp idea,” “Non-profit fundraising ‘experiment,’” “Non-profit, one month pause on any actions and funding,” “Non-profit, agree to terms and resume funding now,” and “B-Corp.” “Greg, Ilya, and Sam all claim the only thing they really care about is there being a long-term control provision that transitions away from short-term majority control at some defined point and ensures a broad group governance of AGI if it’s created,” she writes. Teller responds on October 2nd, asking to “talk through this one live”.

*Exhibit No. 715

An October 20th, 2017 email from Zilis to Teller, describing what appear to be Musk’s views about OpenAI. Musk suggested sticking with a non-profit to “keep clean moral high ground for recruiting,” possibly by looking for funding “with [Bill] Gates types,” though “Tesla solves the funding issue immediately.”

“He thought as people get more scared they will find OpenAI more and timing right for that. My personal take was there is a very small window for that because people generally need to be very scared to take real action, and by the time people are very scared it will probably become nationalized if it’s in the open. Tesla at least has the option to bury,” Zilis concludes. “I will try and get clarity from Elon,” Teller responds. “We need to try a little more to convince Ilya and Greg of this path but then we will be more firm and tell them it’s happening regardless.”

Exhibit No. 716

On October 21st, 2017, Zilis wrote to Teller that OpenAI seems to “clearly prefer the non-profit” over being linked to Tesla. “They think it’s easy to explain why someone would choose OpenAI over Google or Deepmind,” Zilis writes. They also “haven’t internalized the advantages of burying this in Tesla for stealth advantage. They are not naturally hard wired as maneuverers. They have trouble conceiving why this is an advantage.” Teller says they can “stay at the non profit,” and Zilis responds: “Is it useful or not useful to start thinking about how this would work at Tesla, with or without them? Their default will be to put this off as long as possible so if I can help with generating certain types of momentum lmk!”

Exhibit No. 724

A November 27th, 2017 email conversation between Zilis and Tesla’s Sarah O’Brien with “a first stab at an FAQ for thinking through the Tesla AI event.” It describes the creation of a “world class AI lab” at Tesla that would rival Google and Facebook’s labs. It suggests Altman as a possible moderator at the event “to show non-adversarial nature of the OpenAI / TeslaAI relationship? – could be a forcing function for Sam to commit to Tesla AI as well.”

Exhibit No. 728

An December 22nd, 2017 email chain where Zilis forwards Teller an OpenAI status update from Sutskever. Zilis describes a “brief convo earlier this week with Altman about there being zero communication recently and he said he’d try to get things back on track. … I’m not sure if Elon wants in or out on OpenAI, but may be good for him to acknowledge if we are keeping things open?” (She adds, “Sorry if that’s a silly comment.”)

Exhibit No. 753

A February 3rd, 2018 iMessage conversation between Altman and Zilis. “did you think through B Corp subsidiary of Tesla?” Zilis asks. Altman calls it an “interesting idea” that he “had not considered.” Altman then asks Zilis if she plans to go to a SpaceX Falcon Heavy event. “I so badly want to but feels too entitled,” Zilis says, adding a sad-face emoji.

*Exhibit No. 754

A February 3rd, 2018 email conversation between Zilis and Teller. “All signs point to not-Tesla for Greg and Ilya,” Zilis writes. “Altman seems most open and E could probably win him over with a real plan if he wanted to.” Karpathy, she says, “Didn’t fully get why E would want to separate from OpenAI but now seems to mostly understand.”

“Hard to see Greg and Ilya’s egos allowing them to come to Tesla,” Teller says. Zilis disagrees. “I think they likely would come if E *really* did something,” she says. “All they know right now is a vague promise of resources for this and they don’t believe he has a real plan for how to build a counterbalance, which is true because he doesn’t yet.”

Exhibit No. 756

A February 8th and 9th, 2018 email thread between Zilis and Teller. “On AI, do you have any more intel from Greg and Ilya? Are we mainly still trying to get Elon to think about and act on a plan? Recruit Altman? Read stuff from you?” Teller asks. Zilis responds that “They seem pretty set on doing the OpenAI thing. … They all think Elon is an incredible human being but that he really hasn’t done his homework AI / AGI and that really concerns them about working with him.”

Teller and Zilis later agree that, in Zilis’ words, “E just need to have and communicate a little more of a positive vision” about AI.

Exhibit No. 757

A February 11th, 2018 email from Zilis to Musk. “Greg, Ilya, and Altman seem to be leaning toward continuing with OpenAI,” it says, asking if Musk would be open to speaking with them. “I should talk to them this week,” Musk replies the same day.

Exhibit No. 758

A February 13th, 2018 email from Zilis to Musk, offering a series of “thought starters” for how to navigate the future of OpenAI and TeslaAI, including making OpenAI a subsidiary of Tesla and building out TeslaAI by hiring Altman to run it. “Feel free to disregard if this kind of thing isn’t helpful, or if you don’t feel like thinking about AI,” Zilis writes. “The point was merely to try to take some of the weight off of your shoulders for getting to a decision.”

Exhibit No. 766

A February 17th, 2018 iMessage thread between Zilis and Teller, a few days before Musk left the board of OpenAI. It includes strategizing around Musk hoping to “poach 4-5” members of OpenAI. “Should tell OpenAI it’s not a secret. That a couple people may want to come work at Tesla,” Teller says. “Shouldn’t be a covert operation.” Zilis also sends Teller an image macro reading “If you are driving a stolen Tesla, would it be called Edison?”

*Exhibit No. 167

An iMessage thread from February 17th, 2018 between Altman and Zilis. Altman says that with the “number of people elon has reached out to, basically most of openai knows at this point,” saying that he’d been “calling people to try to recruit them,” with the message that “he didn’t think openai could succeed and that he was talking to the 4-5 best people to get them to tesla.”

“This interaction didn’t feel right,” Zilis responds. “I would love to keep helping you as much as I can but important to figure out the correct frameworks of trust for each other.” Altman offers “sincere apologies,” saying that “my goal was simply to give you a heads up,” and they agree to talk in person next week.

Exhibit No. 244

A May 11th, 2019 email from then-OpenAI CTO Mira Murati to Microsoft’s Phil Waymouth. Murati apologises for taking longer than expected to get the “redlines” in an agreement back to him. She mentions the “open source issue” as a sticking point — ”I discussed it more with the team and this continues to be sensitive because we plan to open source the current version and it is not unreasonable that we may continue to open source future versions of the language model,” she writes. She offers six months to a year of exclusivity for Microsoft, as well as the first right of negotiation for commercial licensing of future models.

Exhibit No. 824

An August 27th, 2018 iMessage conversation between Zilis and Reyna Ortiz of Tesla. Zilis asks about scheduling a meeting with Reid Hoffman, saying Altman hadn’t sent documents for Musk to review. She adds: “I am scared of them closing a round without Elon’s approval so I just want to say something to them.”

Exhibit No. 827

An August 31st, 2018 email from Altman to Musk attaching a term sheet for an OpenAI limited partnership — a duplicate of Exhibit No. 236.

Exhibit No. 830

An August 31st and September 1st, 2018 email thread between Zilis, Teller, and Birchall. It offers Zilis’ summary of the term sheet and says “I think it’s critical to ensure Elon is able to come to the right decision for himself on OpenAI now. Once they close $500M raise in this structure it is going to be a big deal and relatively irreversible.”

Exhibit No. 835

A September 9th and 11th, 2018 email conversation between Zilis and Teller with a schedule of Tesla, Neuralink, and other Musk project events. It notes that there is “no longer a need to schedule” a call with Hoffman. “Just for awareness, Elon does not need to do the Reid call because he’s decided to be supportive in spirit of OpenAI but not participate in the new instrument,” Zilis says.

Exhibit No. 877

A July 22nd, 2019 OpenAI blog post titled “Microsoft invests in and partners with OpenAI to support us building beneficial AGI,” announcing a $1 billion investment from Microsoft.

Exhibit No. 900

A July 22nd, 2019 chat thread between Zilis and Altman. “Congrats! I’m sure you’re in touch with him too, but when I’d pinged E the blog post he mentioned he thought it was an impressive deal by you guys,” Zilis says. “Also, you have unreasonably good posture. I need to learn.”

Exhibit No. 925

October 18th, 2020 chat messages from Altman to Zilis. “still think a good idea for me to ping elon for advice on the MSFT fundraise? if so this is about the time for me to do it,” Altman says. “i am positively inclined.”

Exhibit No. 926

An October 19th, 2020 chat conversation between Zilis and Altman. “Yeah tbh I would recommend pinging more often than less on stuff like this as long as it’s not feeling adversarial,” Zilis writes. “If he does give you advice that helps, amazing. Even if not, if he’s cordial he’ll be less inclined to surprise you on Twitter. And if he is rough, it will be the same rough you’ll see on Twitter so at least you’ll know.” Zilis suggests that Musk may “pull the ‘you should have gone with Tesla’ card on you” and suggests Altman find an answer. “ok! thanks,” Altman responds.

*Exhibit No. 302

A document titled “Sept 30, 2022 Feedback from Mira to Sam (only Sam had access to this).” It describes a series of problems on “alignment” between Altman and executives, complaining that “the constant panic around our projects, people, goals etc generates chaos and churn.”

Exhibit No. 1145

A February 15th, 2023 tweet from a user named Genevieve Roch-Decter, CFA reading “Elon Musk says that A.I. is ‘one of the biggest risks’ to civilization and needs to be regulated. He co-founded OpenAI.” Musk responded on February 17th. “OpenAI was created as an open source (which is why I named it “Open” AI), non-profit company to serve as a counterweight to Google, but now it has become a closed source, maximum-profit company effectively controlled by Microsoft. Not what I intended at all.”

*Exhibit No. 1017

A February 25th, 2023 conversation between Zillis and a friend identified as “Shahini Rubicon Fluffer.” Zilis says she has to resign from the OpenAI board because “E’s effort has become well known.” “When the father of your babies starts a competitive effort and will recruit out of openai there is nothing to be done,” Zilis continues. “He proactively apologized that he had pruned my friend network through this.”

*Exhibit No. 301

An undated Slack thread between Murati and OpenAI’s then-general counsel Jason Kwon. As part of a set of questions, Murati asks Kwon about a model referred to as “turbo” (plausibly GPT-4 Turbo, which was announced publicly on November 6th, 2023, shortly before Altman’s removal as CEO) tells Kwon that “Sam told me that you had said it doesn’t need to [go through OpenAI’s deployment safety board (DSB)] per legal. Is this correct?”

Kwon responds: “I actually said something different, which was (1) that people were saying it needed DSB but that also it didn’t seem like that much work and that it would be handled and (2) we should going forward not let anyone just decide something goes through DSB, but instead have a lawyer … be the one to make the determination.” Kwon adds that “Looking at what happened on the turbo thread, it seems like an unhealthy dynamic that the people with the power get to decide when it gets used.”

*Exhibit No. 304

A unanimous consent document from OpenAI’s board of directors dated November 16th, 2023, removing Sam Altman as OpenAI CEO. “The Board has lost trust in Sam Altman’s ability to be candid and forthright in his communications with the Board and employees of the Corporation and OpenAI Global, LLC, and is concerned about the resulting impact of his actions on the Corporation’s mission,” it says, and immediately terminates Altman’s employment at OpenAI, appointing Murati as the interim CEO.

Exhibit No. 1290

A November 17th, 2023 chat message from Zilis to Altman. “Obviously no need to reply but I just wanted to say I hope you are ok. I have no idea what’s going on but you’ve been an awesome person in literally every interaction we’ve had and I care about you as a person first and foremost. Sending all of my positive vibes your way.”

*Exhibit No. 306

A text conversation between Murati and Nadella between November 17th and November 21st of 2023. It asks Nadella urgently to confirm that “Microsoft has assured us that there are positions for all OpenAI employees with the same compensation” if they choose to join. “Satya could you please make a public statement soon that shows support for the joint openai team, basically bringing the team together?” Murati asks. “It’s very important that we don’t lose researchers to Demis or Elon. The technical team is being dragged in so many recruiting directions and a unified front would help immensely.”

Exhibit No. 1048

A duplicate of Exhibit No. 306.

*Exhibit No. 309

A text conversation between Murati and a partially redacted contact referred to as “Kevin” (likely Microsoft CTO Kevin Scott) between November 17th and November 21st, 2023. Murati says that they are “Close to having the board resign,” noting “Ilya signed our petition.” Kevin asks for the text of the petition and says “Shockingly given how close everyone is covering this right now, I haven’t heard mention of it, other than through OAI->Microsoft employee conversations.”

*Exhibit No. 1046

An undated copy of the petition referred to above, calling on OpenAI’s board of directors to resign, saying “Your actions have made it obvious that you are incapable of overseeing OpenAI.” It includes the comment that Microsoft has ensured positions for all OpenAI employees and has 751 signatories.

*Exhibit No. 315

A text conversation between Altman and Murati dated November 19th and November 20th, 2023. Altman asks “can you please officially invite me to the office for a meeting?” Murati agrees and asks for updates between what appear to be Altman’s talks with Microsoft. “can you indicate directionally good or bad? satya and others nervous,” Altman says. “Directionally very bad,” Murati responds. The pair of them talk about the board’s decision to remove him. “New guy is rando twitch guy. They don’t want you,” Murati says — referring to Twitch cofounder Emmett Shear, who was briefly appointed CEO.

Altman ends the conversation saying “i think you all just need to get a petition of everyone saying they will quit and join.”

Documents released May 7, 2026

Exhibit No. 222

A tax exemption application for OpenAI dated September 1st, 2016.

Exhibit No. 22

An August 28th, 2017 registration of a charitable trust in the state of California for OpenAI, including copies of its initial incorporation documents, bylaws, and establishment as a tax-exempt organization.

*Exhibit No. 229

An August 11th, 2017 to January 12, 2018 email conversation between Nadella, Altman, and several other figures at Microsoft. It begins with Nadella congratulating Altman on the outcome of an early win in OpenAI’s Dota 2 project. Altman proposes a “big partnership” on the next phase of the project, saying it could “lead to major new breakthroughs in AI but will require huge amounts of compute, probably something like $300MM at Azure list prices.”

Executive Jason Zander mentions past concern that “we don’t want ‘machines beating humans’ and are not supportive of any push on this,” but Microsoft and OpenAI enter negotiations anyway. Over the course of internal discussion, Scott says that “I’m highly skeptical of an imminent breakthrough in AGI” and complains that Microsoft hasn’t gotten much out of its existing deal with OpenAI — “they’re treating us like a bucket of undifferentiated GPUs, which isn’t interesting for us at all.” But he expresses concerns about the “PR downside of us not funding them, and having them storm off to Amazon in a huff and shit-talk us and Azure on the way out.”

The thread ends with Altman apparently indicating that “he would start looking for alternative solutions for capacity” as the conversation about the deal continues. There’s also an attached copy of what’s apparently an earlier presentation on OpenAI and Azure.

Exhibit No. 53

OpenAI’s 2018 tax returns as a 501(c)(3) tax-exempt organization. They list $50 million in revenue for the year and $51.5 million in total expenses, and note that OpenAI released its charter that year. Accomplishments include “scaling its reinforcement learning algorithms to beat a team of 99.95th percentile Dota 2 players,” and it mentions that Brockman and policy director Jack Clark testified before Congress “to discuss the importance of developing shared ethical norms, the need to support AI development & education, and the need for government-led measurement and forecasting of AI.”

Exhibit No. 54

OpenAI’s nonprofit tax returns for 2019, the year it established a for-profit arm. The returns list $33.6 million in contributions and grants and $3.3 million in total expenses. Its achievements include developing the GPT-2 model, which among other developments “are moving the Organization closer to achieving its mission, which is the development of Artificial General Intelligence in the public interest.”

Exhibit No. 55

OpenAI’s 2020 nonprofit tax returns. They list $3.5 million in revenue and $13 million in expenses and note that “through its control of … a capped-profit company to rapidly scale investments in compute and talent,” it introduced GPT-3, Image GPT, and musical neural net Jukebox, among other achievements.

Exhibit No. 56

OpenAI’s 2021 nonprofit tax returns. They list roughly $11,700 in revenue, $1.4 million in expenses, and accomplishments including the DALL-E image generator, the Codex code generator, and “WebGPT,” the precursor to ChatGPT.

Exhibit No. 57

OpenAI’s 2022 nonprofit tax returns. They list around $44,500 in revenue, $1.3 million in expenses, and accomplishments including DALL-E 2, “new and improved content moderation,” and ChatGPT.

Exhibit No. 1040

An OpenAI blog post from November 17th, 2023 titled “OpenAI announces leadership transition” announcing Altman’s departure and Murati’s appointment as interim CEO.

*Exhibit No. 319

A November 20th, 2023 email (during Altman’s short ouster) from Toner to Sutskever, the contents of which are a Slack message “from the whole board to the team.” It says the OpenAI board “firmly stands by its decision … Put simply, Sam’s behavior and lack of transparency in his interactions with the board undermined the board’s ability to effectively supervise the company in the manner it was mandated to do.”

The board says it has been meeting with employees, investors, and others and that new CEO Emmett Shear will soon say more publicly.

Exhibit No. 1049

A duplicate of Exhibit No. 319.

Exhibit No. 58

OpenAI’s 2023 nonprofit tax returns. They list $5.4 million in revenue, $2.9 million in expenses, and accomplishments including sponsoring a “comprehensive basic income study,” “experimenting with education-centered programs like OpenAI Scholars,” and developing GPT-4 and ChatGPT.

Exhibit No. 284

An October 3rd, 2025 court declaration from Robert Wu, OpenAI’s deputy general counsel, about how OpenAI evaluated target redemption amounts (TRAs) for employees and investors who received economic interests in OpenAI’s for-profit branch.

Documents released May 11, 2026

*Exhibit No. 220

An email chain between Nadella and Altman between March 29th and April 17th, 2015. Altman tells Nadella that he and Musk are drafting a letter to the US government, asking for a new AI safety agency, and asks if Nadella can sign. “I think this is probably the biggest risk to the continued existence of humanity that most people are ignoring,” he says.

After saying he’s “interested” in supporting the letter, Nadella later follows up. “A lot of our folks feel this is really premature,” he writes. “Issue of human safety and the control problem will become real issues, but I think the call here needs to for federal funding and encouragement of research vs call of regulatory invention.” Altman says he’s changing the letter based on the response — it’s not clear whether it was finalized or sent.

Exhibit No. 63

A November 18th, 2015 email from Altman to Musk discussing Sutskever. “Ilya is trying to decide between leading YC AI and a massive counteroffer from Google. I think talking to you would help him make the right choice :)” Altman writes. Sutskever responds, confirming that he’d like to speak with Musk. “No problem, will call today,” Musk writes.

Exhibit No. 221

A December 12th, 2015 email from Nadella to several other people from Microsoft. It includes a link to OpenAI’s introductory blog post. “Did we get called to participate?” Nadella asks. “AWS seems to have sneaked in there.”

Exhibit No. 228

The August 11th, 2017 congratulatory email from Nadella contained in Exhibit No. 229, as well as responses from Brockman, Altman, and Musk. “Thank you. Used quite a lot of Azure to make it happen :),” Brockman says. “thank you satya! we really appreciate the partnership with you guys,” says Altman. Musk responds: “Indeed, much appreciated. Will make sure people know about Microsoft’s help.”

Exhibit No. 1571

A February 27th email conversation between Altman and Nadella discussing a March 13th OpenAI meeting. It notes that “Elon recently departed the OpenAI board, though is still advising us” and describes plans for a fundraising round potentially involving a “coalition of non-Google tech companies.” Nadella, Altman, and other OpenAI figures agree to meet.

*Exhibit No. 230

A March 11th, 2018 email chain between several Microsoft figures, including Nadella and Scott, discussing plans for the future of Microsoft’s deal with OpenAI. Near the end of the thread, Scott says the deal “seems like something that in success (which by no means is guaranteed) would be competitive.” “One of the big questions” he has for Altman is “whether or not they intend to make the hardware and software open source in the original spirit of OpenAI. IMO that would be good competitive insurance and something that might be worth funding. I wonder if the big OpenAI donors are aware of these plans? Ideologically, I can’t imagine that they funded an open effort to concentrate ML talent so that they could then go build a closed, for profit thing on its back.”

Exhibit No. 245

A June 10th, 2019 memo from Hood and Scott to Microsoft’s board of directors, seeking approval to invest $1 billion in OpenAI. Even if OpenAI “never achieves full AGI, the attempted effort has the likely potential to produce a variety of positive interim results that can have high commercial value,” it says.

Exhibit No. 257

A February 5th, 2022 email from Altman to Scott and Microsoft’s Chris Young, where Altman says OpenAI is “working on a restructure” and wants to run it by them. Young and others agree to the meeting, and on February 13th, Young emails Hood and Microsoft deputy general counsel Keith Dolliver. “At a high level, it all seems to make sense and they assured us this doesn’t really change anything related to our interests,” Young writes, but he says Hood and Dolliver will likely “want to dive deeper on this with them.”

*Exhibit No. 259

An April 28th, 2022 email from Nadella to several figures at Microsoft, including Dolliver. It discusses how to continue the OpenAI partnership “in a more aligned way.” In a series of notes, Nadella says in exchange for funding, “we want full IP rights and also need to ‘embed’ our folks across every layer of the stack.” “The biggest issue I see is all these investors. I just don’t get their incentives,” Nadella continues. An attached document contains a description of the plan for a deal. It begins “TL;DR: Microsoft funds OpenAI sufficiently to stay in the lead for AGI. Microsoft gets perpetual exclusive access to OpenAI IP as long as that happens (i.e., if Microsoft chooses to stop, then go-forward exclusivity stops).”

*Exhibit No. 261

A July 11th, 2022 email from Altman to Nadella, Scott, and others. Altman describes two possible paths to a deal: one in which “We stick with the current structure … but make it so that any MSFT team can easily use the models,” and another that Altman calls “the ‘really go for it’ deal.” That deal involves Microsoft investing in OpenAI heavily and getting “all IP except hyperscale training code and except where there is a legitimate AGI safety issue.” Microsoft would get all net revenue generated by OpenAI, and they’d create a joint infrastructure team for hardware.

Nadella forwards the email internally and discusses it with others at Microsoft, saying he has reservations about the first option. “‘Hyperscale training code’ is not a thing,” adds Mikhail Parakhin, then president of web experience. “I think we just need to insist on all IP.” Hood says she believes “we have to opt out of the business of funding their capex upfront.” The thread ends with Nadella saying that “I want to spend this money,” but “if we are going to spend this kind of money and not have control of destiny, it makes no sense.”

Exhibit No. 267

A January 11th, 2023 memo from Nadella, Smith, Hood, and others to Microsoft’s board of directors seeking an additional $10 billion investment in OpenAI on top of the existing $3 billion deal. It outlines the terms of the investment, which it notes would be made as OpenAI restructured and converted to a for-profit company.

Exhibit No. 269

A text conversation between Altman and Nadella on January 14th, 2023. Nadella asks when OpenAI will activate paid subscriptions for ChatGPT, and Altman says it’s aiming to be ready by the end of January but asks if Nadella has a preference. “Overall getting this in place sooner is best. At some level Bing is the scoped ad supported version,” Nadella says. “So we collectively have both in market on top of the same base model.” They agree to run through the ChatGPT roadmap together when Nadella returns from Davos.

Exhibit No. 270

A January 15th, 2023 text conversation between Altman and Nadella. It continues the conversation thread about ChatGPT paid subscriptions, with Nadella saying “end of Jan will good.”

Exhibit No. 272

A January 29th, 2023 text conversation between Altman and Nadella, who asks how many subscriptions “have you guys added to ChatGPT.” Altman says “we are at 6mm DAU which is our capacity limit. have 50mm that have signed up ever but have to turn away.” He notes that paid subscriptions were delayed “due to legal issues” but should go out soon.

Exhibit No. 305

The November 17th, 2023 OpenAI blog post about “leadership transition” and Altman’s firing that’s also provided in Exhibit No. 1040.

*Exhibit No. 311

A November 18th and 19th, 2023 series of texts between Nadella and Altman around Altman’s firing. Nadella offers “one idea” and asks Altman to call him, then shares a message from “Brad” (likely Microsoft President Brad Smith). “If it’s needed, we’ll have a new subsidiary opened on Monday,” it says, tentatively calling the endeavor “Microsoft RAI, Inc.” From there, it promises, Microsoft will “support Sam in whatever way is needed.”

Exhibit No. 313

November 19th, 2023 messages between Altman, Nadella, and Taylor. It discusses the composition of a board of directors, with Altman saying that “we need stability right now” and expressing concerns with adding “someone new we don’t know.” Nadella agrees that “I would hate for us to stay in this state of instability and letting our team down, giving our competitors a field day and ultimately letting the mission down as well.”

*Exhibit No. 316

A November 18th to 20th, 2023 email thread between OpenAI’s Hannah Wong and several people at Microsoft, including Nadella, Smith, Hood, and Scott. Wong sends the draft of a Slack post about Altman’s firing, saying that the OpenAI board had resigned and that Hood would become an observer to it in a non-voting capacity. Microsoft chief communications officer Frank Shaw says that “we’re going to need to drive a set of stories with OAI that establishes a clear narrative of what happened … The coverage and story so far is all over the map.”

Scott then lays out a timeline of events, beginning with Sutskever being “increasingly at odds” with Altman. It describes Ilya as being concerned by a “perfectly natural tension” between the research and applied divisions of OpenAI, among other problems, and convincing the board to oust Altman instead of appealing to him like he would “in a normal company.”

“Two of the board members were effective altruism folks who all things equal would like to have an infinite bag of money to build AGI-like things, just to study and ponderk but not to do anything with,” he adds. “None of them were experienced enough with running things, or understood the dynamic at OpenAI well enough to understand that firing Sam not only would not solve any of the concerns they had, but would make them worse.”

Exhibit No. 317

A November 20th, 2023 X post by Nadella. “We remain committed to our partnership with OpenAI,” it begins, and announces that “Sam Altman and Greg Brockman, together with colleagues, will be joining Microsoft to lead a new advanced AI research team.”

Exhibit No. 318

A November 20th, 2023 chat conversation between Lightcap, Altman, and Nadella. It discusses a “rewrite” of an unpublished statement, and Altman suggests saying that “satya and my top priority remains to save openai … our partnership makes this very doable.” Nadella suggests modifying the top priority to be “ensure OpenAI continues to thrive… that will make it more positive.” Lightcap calculates how much equity would need to be issued to employees and says it could range from $25 billion to $29 billion, depending on Sutskever’s equity.

Exhibit No. 320

An extensive November 20th, 2023 chat conversation between Nadella, Scott, Smith, and Altman. “have a positive update, ish,” Altman says, saying Shear thinks “its looking reasonably positive for the 5 member board option.” They apparently discuss a series of potential board members including economist Larry Summers, PayPal board member Belinda Johnson, and Coinbase president Emilie Choi, among many others. “excited to get this all sorted,” Altman says.

Exhibit No. 1057

A December 1st, 2023 OpenAI director certificate with an attached copy of resolutions adopted November 29th, including Altman’s reappointment as CEO and a reorganization of the board after Sutskever, Toner, and Tasha McCauley resigned. Summers and Sierra CEO Bret Taylor are appointed instead.

Exhibit No. 1068

Minutes of an April 23rd, 2024 mission and strategy committee of the OpenAI board of directors. It discusses plans for changing OpenAI’s capped-profit structure and the challenges of shifting away from nonprofit control, including “how control by the Company is perceived by policymakers, regulators, employees, recruits, press and other public stakeholders.” OpenAI would officially announce plans to become a public benefit corporation at the end of 2024.

Exhibit No. 1150

A December 27th, 2024 OpenAI blog post titled “Why OpenAI’s structure must evolve to advance our mission,” outlining plans for a public benefit corporation that would control OpenAI’s operations and business, relegating the nonprofit to a smaller role.

Exhibit No. 279

A September 9th, 2024 memo from Nadella, Scott, and others to OpenAI’s board of directors proposing an investment of $1 billion in OpenAI’s convertible financing. It says OpenAI has reached “notable business milestones,” including 350 million monthly active users and 175 million weekly active users for ChatGPT, 10 million paid subscribers, and annualized revenue growth “of 3.4x to $3.6b billion as of June 2024.”

Exhibit No. 1165

Approval of minutes from a January 14th, 2025 OpenAI board of directors meeting, signed by board members between late May and mid-June of that year. The meeting discussed negotiations with Microsoft over its restructuring plan, dubbed “Project Watershed”. The overarching objectives were “ensuring [the restructured OpenAI] receives fair value in Project Watershed” and securing its ability to keep developing OpenAI.

Exhibit No. 1163

An OpenAI blog post from May 5th, 2025 titled “Evolving OpenAI’s structure.” It describes updates to its plan that would still found a public benefit corporation, but one that “will continue to be overseen and controlled by nonprofit” and have a large stake in the for-profit arm. The move was widely interpreted as backing off its earlier for-profit plans, following regulatory concerns and public protest from figures including Musk.

Exhibit No. 1528

A memo from Hood and Scott to the Microsoft board of directors on June 10th, 2019, discussing forming a partnership with OpenAI’s for-profit subsidiary and seeking approval for an investment of $1 billion.

Exhibit No. 1168

An OpenAI blog post from September 11th, 2025 titled “Statement on OpenAI’s Nonprofit and PBC” concerning the equity stake of OpenAI’s nonprofit in the newly organized public benefit corporation.

Exhibit No. 1597

A September 29th, 2025 X post by Nadella welcoming Musk’s AI model Grok 4 to Microsoft’s Azure AI foundry. “Thanks Satya!” Musk says in a repost.

Documents released May 12, 2026

Exhibit No. 3

A May 25th, 2015 email from Altman to Musk. “Been thinking a lot about whether it’s possible to stop humanity from developing AI. I think the answer is almost definitely not,” he says. “Any thoughts on whether it would be good for YC to start a Manhattan Project for AI?” Musk responds: “Probably worth a conversation.”

Exhibit No. 23

An April 2nd, 2018 email from Altman to Musk with a draft of the OpenAI Charter. “We are planning to release this next week–any thoughts?” Altman asks. “Sounds fine,” Musk responds.

Exhibit No. 253

A February 10th to February 24th, 2021 email conversation involving Altman, Murati, Scott, and others, about new investment terms. Altman reiterates multiple times how worried OpenAI is about sharing its models with Microsoft employees based in China.

*Exhibit No. 107

A February 18th, 2023 text conversation between Musk and Altman. Altman says he recalls a TV interview where Musk described “being attacked by some guys, and you said they were heroes of yours and it was really tough. well, you’re my hero and that’s what it feels like when you attack openai. totally get we have some screwed some stuff up, but we have worked incredibly hard to do the right thing, and i think we have ensured that neither google nor anyone else is on a path to have unilateral control over AGI, which i believe we both think is critical.” He thanks Musk for his help and continues: “i don’t think openai would have happened without you—and it really fucking hurts when you publicly attack openai.”

“I hear you and it is certainly not my intention to be hurtful, for which I apologize, but the fate of civilization is at stake.” Altman says he’d love to hear suggestions, but it’s “not clear to me how the attacks on twitter help the fate of civilization.” He adds that he’s spoken to others at OpenAI about whether they’re recruiting from Tesla, and he assures Musk that “i will make sure we don’t hurt tesla, i obviously think it’s a super important company.”

*Exhibit No. 312

A November 19th to 20th, 2023, text conversation between Altman, Nadella and Bret Taylor, during the weekend of Altman’s ouster. It paints a picture of the scramble between different plans and how things changed minute to minute. They mention chatting with an Adam, presumably board member Adam D’Angelo, and discuss a potential new three-person board for the company, likely with Microsoft CFO Amy Hood as observer and Taylor as board chair. Taylor gives his requirements for accepting such a position.

Altman writes at one point, “ok i think we get ready to go on the plan of mira rehiring me and greg while we work on the injunction. the board just won’t give any timeline. this will stabilize things in the short term and everyone can come to work monday morning. ok with you?” Altman then adds, “if that blows up we can go on to the subsidiary.”

*Exhibit No. 258

A February 2022 email chain, in which Altman starts off asking Microsoft CTO Kevin Scott and Microsoft business development chief Chris Young if they have time to chat about OpenAI’s restructuring. OpenAI’s Lightcap and Kwon are CC’d.

“We are working on a restructure to OpenAl to address some of the challenges we’ve faced with our unusual structure (a nonprofit being in control of an LLC) as we become a more commercial effort,” Altman writes.

Lightcap asks for expedited approval from Microsoft about the restructuring. He writes, “We normally wouldn’t ask for such a tight turnaround, but we’re back-solving for executing by 6/30. The reason is new CA legislation that goes into effect on 7/1 that would require the AG to review the transaction given it involves a nonprofit.”

*Exhibit No. 720

A November 2017 text conversation between Altman and Sam Teller, Musk’s chief of staff. Teller tells Altman that he (Altman) and Musk should talk.

“Regarding long-term control, he’s not super concerned. The Tesla board or shareholders can always boot him (he owns just 22%) and the government can always intervene if if it gets to the point where that is necessary. As far as maintaining the moral high ground in recruiting, there’s no way to stay as pure as the non-profit, but I think we can pitch the same OpenAl fundamental vision – that Al should be safe and beneficial for humanity – in contrast to the sketchy, indecipherable principles of DeepMind. We can do this at the same time that we communicate the benefits of Al for self-driving cars, energy grid optimization, etc.

Teller adds that Musk “remains very open to the idea” of Altman joining the Tesla board, and that he’d like for OpenAI to still exist, just for people remaining there to purely do research or to split time between OpenAI and Tesla. Regardless of how the conversations about OpenAI and Tesla go, Teller says, Musk is “committed to building a stronger Al team within Tesla for both hardware and software.”

“If we could have you involved, and hopefully Greg and Ilya, combined with everything that Elon and Tesla bring to the table, I think that will give us the greatest odds of success in rapidly building AGI (that doesn’t obliterate us),” Teller writes.

Exhibit No. 808

An April 2018 text conversation between Altman and Zilis. Zilis called the meeting earlier Musk’s “happiest most calm meeting by far.” She mentions that she’d like for Altman and Musk to meet about “the investment thing,” though, so that it won’t “irk” Musk later on.

*Exhibit No. 755

A February 2018 text conversation between Teller and Zilis. They discuss which of OpenAI’s co-founders — particularly Altman, Brockman, and Sutskever — they’d like to recruit to Tesla, and the pros and cons of each. (For instance, Brockman is “good for recruiting [and] getting shit done,” Teller writes, while “Ilya [is] brilliant but seems least central.”) Zilis writes that Sutskever took a “massive pay cut” when Musk stopped funding OpenAI.

“Sam is winnable if E wants,” Teller writes. “He clearly doesn’t want to live a life without Elon if possible.” Zilis said that it’d be a good idea to allow Altman to watch a SpaceX launch “to keep the love flowing.”

“I think if you win altman you force their hand,” Zilis writes. “But if I do think altman stays it’s very real.”

Teller says, “I don’t love OpenAI continuing without Elon. Would rather disable it. By recruiting the leaders. It can be a little research thing. But we should try to suck up the top talent. To Tesla.”

Zilis says that if Musk doesn’t enter the space with a competitor, she disagrees, since “we want this [technology] in the world,” but if he does choose to compete, then she agrees.

Exhibit No. 770

A February 20th, 2018, press release announcing OpenAI’s “new donors” and announcing that Musk will depart the OpenAI board.

*Exhibit No. 648

An August 2017 email conversation between Altman, Teller, and Zilis.

“I desperately want to see this work with Elon,” Altman writes. “I think it is critical that our mission be fulfilled, and Elon is an important part of that. I have spent a lot of time talking to Greg and Ilya in the past few days about why I believe this is so critical.”

Altman adds that Brockman and Sutskever aren’t “really feeling respected” by Musk right now and asks Teller and Zilis to “help Elon communicate with them differently.”

“Speaking for myself, I’m not very sensitive to economics (I think if we succeed, money will not matter) but I am worried about control,” Altman writes. “I don’t think any one person should have control of the world’s first AGI–in fact, the whole reason we started OpenAl is so that wouldn’t happen. I’m open to creative structures, though, and I’m less sensitive to intermediate-term control.”

Teller responds, “Elon isn’t asking for absolutely eternal power, but he needs to be able to make critical and often counterintuitive company decisions when push comes to shove. That’s the only non-negotiable for him. It is nothing personal – just something that will be true of every company he starts for the rest of his life.” But Teller adds that Musk would likely be “open to creative longer term structures” and that they can brainstorm.

Exhibit No. 1305

A graph showing contributions to OpenAI, Inc. between 2016 and 2023.

Exhibit No. 928

A November 2020 text conversation between Altman and Zilis. Altman asks for feedback on his latest meeting with Musk, and Zilis relays a “generally positive sentiment.”

Exhibit No. 861

A March 2019 text conversation between Altman and Zilis. Altman asks Zilis to tell Musk that OpenAI plans to announce a new structure and funding soon and that “at this point” Altman is planning to tell Musk about the potential Microsoft investment, since “it’s getting serious.”

Exhibit No. 864

A March 2019 text conversation — a follow-up to the above conversation — between Altman and Zilis. Altman says Musk told him that even though official language states Musk isn’t formally involved with OpenAI, he’d “like it if we say on background that he’s still helpful.”

Exhibit No. 1190

A certificate of incorporation for OpenAI’s public benefit corporation, signed by Altman on October 28th, 2025.

Exhibit No. 876

A July 2019 document prepared by Hemming Morse, detailing OpenAI’s valuation: about $60.380 million.

Exhibit No. 1306

A chart of contributions to OpenAI between 2016 and 2023. It says Alameda Research, the cryptocurrency trading firm founded by Sam Bankman-Fried, contributed $500,000 in 2018.

Exhibit No. 385

An incremental update in Musk v. Altman, including Altman’s supplemental responses and objections to some of Musk’s claims. Much is redacted.

Documents released May 13, 2026

Exhibit No. 1304

A graph showing Musk’s monetary donations to OpenAI.

*Exhibit No. 957

An October 2022 email chain in which Altman sends OpenAI board members a Microsoft term sheet and gives other updates. Toner responds with a list of detailed questions. She writes, “I think there’s a real possibility that 5 or 10 years from now, people look back and think of the main role OpenAl played during the late 2010s/early 2020s as being the org that set off great excitement about & investment in AGI (and then lost its lead to other orgs). At a minimum, that would be a shame for OpenAl; at a maximum, it could mean that OpenAl’s main legacy is to have amped up the race to AGI and thereby caused significant harm. I see the primary challenge here as how to balance risks of that dynamic against the potential gains of releasing new models in one form or another.”

In a response, McCauley writes that the term sheet Altman sent references the OpenAI restructure plan and that she’d like to talk about the governance implications. “The delay on the administrative restructure gave us more time to consider the governance restructure, but we still haven’t seen any draft of legal language for the change to the Nonprofit’s governance rights,” she writes. “I’d like to make sure that any steps we take toward putting a corporation into the structure don’t lock us into a new set of governance rights (or make it very difficult/ expensive to change course) before we’ve had a chance to fully understand and approve.”

Exhibit No. 1532

A February 2019 document detailing OpenAI’s restructuring, signed by Altman and Holden Karnofsky, a former board member at OpenAI (who has since departed for Anthropic, and is also married to Anthropic co-founder Daniela Amodei).

*Exhibit No. 1516

A November 2018 email from Microsoft CTO Kevin Scott to Nadella. He details his takeaways from a dinner with Altman and others, including a lot of aspects of OpenAI’s compute and current operations. He says that OpenAI’s “rate of change is accelerating. They probably made more progress last year than Deep Mind, even though Deep Mind out mans them 10:1 in terms of PhDs.” He also says that Google is “going absolutely nuts chasing reinforcement learning talent” and that they just made a cash offer to an OpenAI engineer for $20 million per year.

OpenAI’s “next ‘stunt,’” Scott writes, “is that they are now able to generate novel, short video content that is indistinguishable from something made by humans. This isn’t deep fakes; it’s 60 second novel video content. They now believe that they’re 50% likely to achieve AGI by 2030.”

Exhibit No. 1534

A collaboration agreement between the OpenAI nonprofit and the OpenAI LP, dated October 10th, 2018.

Exhibit No. 1531

A legal document detailing OpenAI’s restructuring.

Exhibit No. 1601

OpenAI’s certificate of limited partnership, dated September 2018.

Exhibit No. 1600

An extensive internal memo from March 2019, sent to Microsoft’s board of directors by CTO Kevin Scott and CFO Amy Hood, with the subject line “Investing in the Future of Large Model Artificial Intelligence.”

It details how Microsoft can participate in the AGI chase by way of OpenAI’s research, and that the company “would have a stake in the commercial returns on AGI, should they achieve it.”

*Exhibit No. 1598

A March 2018 email chain between Microsoft executives, discussing OpenAI, its prospects, and whether they should invest. Brett Tanzer says that the OpenAI team “estimate[s] that they will need from $5-10B of compute over the next 3-4 years, much of the capital is needed in years 3-4 vs early on.”

He also details OpenAI’s plan to achieve AGI, particularly that “they have a recipe in mind to breakthrough on AGI that they describe as: One Al Learning Algorithm + Burst of Hardware Speedup + Competitive Multi Agent Training Environment.”

Exhibit No. 1604

A consolidated balance sheet for OpenAI, comparing 2018 and 2017.

Exhibit No. 1602

An amended conflict of interest policy for OpenAI, dated February 2019.

Exhibit No. 1603

A capitalization table for OpenAI, seemingly from 2019.

We translated the Palantir manifesto for actual human beings

2026-04-21T18:39:04-04:00

Palantir CEO Alex Karp is a man in charge of one of the most important and frightening companies in the world. Karp’s new book, cowritten with Nicholas Zamiska, is called The Technological Republic. After claiming “because we get asked a lot,” Palantir posted a 22-point summary of the book that reads like a corporate manifesto. It evokes both weird reactionary shit and also trilby-wearing Reddit comments from the early 2010s.

Palantir’s summary of the book is ominous. But even the company’s name is unironically ominous. The palantíri are crystal balls in The Lord of the Rings that let Middle-earth’s worst tyrants spy on the heroes of the story. It’s a fun reference if you have no shame about your company’s mission.

We’ve attempted to translate these 22 points from Alex Karp’s alien words into something more reasonable, like human words from someone who might play him in the biopic. (Hello, Taika Waititi.) In so doing, we’ve become much more sympathetic to why Jürgen Habermas refused to supervise Karp’s research.

1. Silicon Valley owes a moral debt to the country that made its rise possible. The engineering elite of Silicon Valley has an affirmative obligation to participate in the defense of the nation.

Translation: Silicon Valley has an enormous opportunity to extract as much money from federal government defense contracts as possible. To do this, we will bring back a draft for engineers. We’re really into bringing back the draft. Deepfaked teenagers, low-paid gig workers, and victims of the Rohingya genocide need not apply.

2. We must rebel against the tyranny of the apps. Is the iPhone our greatest creative if not crowning achievement as a civilization? The object has changed our lives, but it may also now be limiting and constraining our sense of the possible.

Translation: We can’t say “we wanted flying cars, instead we got 140 characters” anymore because Elon Musk lets you write essays on Twitter now. Though if you thought the apps were tyrannical, wait until you get a load of us.

3. Free email is not enough. The decadence of a culture or civilization, and indeed its ruling class, will be forgiven only if that culture is capable of delivering economic growth and security for the public.

Translation: People are mad at tech billionaires for their obscene wealth and arrogance. Instead of winning them over by providing free access to a useful everyday service, we’re gonna sell a lot of software that will let the government spy on them while demanding tax cuts.

4. The limits of soft power, of soaring rhetoric alone, have been exposed. The ability of free and democratic societies to prevail requires something more than moral appeal. It requires hard power, and hard power in this century will be built on software.

Translation: Words and feelings are free, which is why we want to sell weapons. Nobody got rich suing for peace.

Trust in a CEO who studied the blade:

5. The question is not whether A.I. weapons will be built; it is who will build them and for what purpose. Our adversaries will not pause to indulge in theatrical debates about the merits of developing technologies with critical military and national security applications. They will proceed.

Translation: “Soft power” and “ethics” are beta shit for Broadway shows and Dario Amodei. Hear that, Pete Hegseth? We’re warriors — pay up.

But seriously. If our enemies have no oversight then why should we? The future is an AI battlefield and we need rules of engagement that let us cook. Which is to say: Forget the rules of engagement. The government is not coming to save you — we are. The world is too dangerous for us to be governed by the law of armed conflict.

Welcome to the 21st century: safety not guaranteed.

6. National service should be a universal duty. We should, as a society, seriously consider moving away from an all-volunteer force and only fight the next war if everyone shares in the risk and the cost.

Translation: We’re going to bring back the draft. Our vision of permanent war only works if we courageously volunteer people 40 years younger than us to die for oil.

7. If a U.S. Marine asks for a better rifle, we should build it; and the same goes for software. We should as a country be capable of continuing a debate about the appropriateness of military action abroad while remaining unflinching in our commitment to those we have asked to step into harm’s way.

Translation: Sure, those wimps at Anthropic are selling an AI system they claim has spotted cybersecurity vulnerabilities in “every major operating system and web browser.” But Pete, seriously: We will kill anybody you want with our software guns.

8. Public servants need not be our priests. Any business that compensated its employees in the way that the federal government compensates public servants would struggle to survive.

Translation: We care about wages – which is why we think Washington’s revolving door of lobbying and office-holding should be way more lucrative for everyone. There are mountains of cash for people who will look the other way.

And if you’re not on board? Well, all those pesky bureaucrats who do things like “investigate fraud” and “enforce safety standards” and “administer the social safety net” are holier-than-thou myrmidons who should be fed into the DOGE wood chipper.

9. We should show far more grace towards those who have subjected themselves to public life. The eradication of any space for forgiveness—a jettisoning of any tolerance for the complexities and contradictions of the human psyche—may leave us with a cast of characters at the helm we will grow to regret.

Translation: If you made fun of that video where our CEO looks like he’s on cocaine, you’re responsible for the rise of fascism. Also, we’re going to be conveniently vague about what “those who have subjected themselves to public life” means, because “be nicer to multimillionaires who go on podcasts” doesn’t have the same ring. Oh, and if you complain about the IT Renfields of DOGE, you’re anti-American.

10. The psychologization of modern politics is leading us astray. Those who look to the political arena to nourish their soul and sense of self, who rely too heavily on their internal life finding expression in people they may never meet, will be left disappointed.

Translation: Society must stop centering sensitive crybabies who want to feel personally validated by elected officials and filter their politics through emotional reactions. Also, I feel strongly that Zohran Mamdani is a pagan who is going to Wicker Man me.

11. Our society has grown too eager to hasten, and is often gleeful at, the demise of its enemies. The vanquishing of an opponent is a moment to pause, not rejoice.

Translation: Your quote-dunking on that video of our CEO yelling “I’m sure you’re enjoying this as much as I am!” while bragging about how Palantir must “scare our enemies and, on occasion, kill them’” was snide, uncalled for, and frankly crass.

12. The atomic age is ending. One age of deterrence, the atomic age, is ending, and a new era of deterrence built on A.I. is set to begin.

Translation: History has rendered absolutely no complex or ambivalent judgments upon the nuclear arms race, so let’s go ahead and repeat it. Why spend money making sure nukes don’t explode by accident when you can fund AI instead? The atomic bomb is so last century.

13. No other country in the history of the world has advanced progressive values more than this one. The United States is far from perfect. But it is easy to forget how much more opportunity exists in this country for those who are not hereditary elites than in any other nation on the planet.

Translation: We canceled our internal DEI programs but we’re fully prepared to steal valor from everyone in US history who fought to make it a more perfect union.

14. American power has made possible an extraordinarily long peace. Too many have forgotten or perhaps take for granted that nearly a century of some version of peace has prevailed in the world without a great power military conflict. At least three generations — billions of people and their children and now grandchildren — have never known a world war.

Translation: Si vis pacem, para bellum, baby! We’ll conveniently leave out all of the regional and secret wars the US has engaged in over the years or the fact that Trump recently derailed the world economy by launching a war of aggression after campaigning on a promise of no new wars. We will not elaborate on what “next war” Point Six was talking about.

15. The postwar neutering of Germany and Japan must be undone. The defanging of Germany was an overcorrection for which Europe is now paying a heavy price. A similar and highly theatrical commitment to Japanese pacifism will, if maintained, also threaten to shift the balance of power in Asia.

Translation: We can definitely sell software to a militarized Germany and Japan too!

16. We should applaud those who attempt to build where the market has failed to act. The culture almost snickers at Musk’s interest in grand narrative, as if billionaires ought to simply stay in their lane of enriching themselves . . . . Any curiosity or genuine interest in the value of what he has created is essentially dismissed, or perhaps lurks from beneath a thinly veiled scorn.

Translation: Elon Musk is our sin eater but I’ll be damned if I let a fellow billionaire loser get dunked on so much for his posts on main. Also, if you raise too many doubts about his IPO, my friends and I are going to lose a lot of money.

17. Silicon Valley must play a role in addressing violent crime. Many politicians across the United States have essentially shrugged when it comes to violent crime, abandoning any serious efforts to address the problem or take on any risk with their constituencies or donors in coming up with solutions and experiments in what should be a desperate bid to save lives.

Translation: Federal government money is nice, but are we tapping state and local? Sure, politicians talk constantly about violent crime and it’s been on a huge downswing statistically over the past decade, but that’s not what we’ve been seeing in our local “VC lives matter” group chats. No further questions about who’s doing the violent crime these days, please. Get those Ring cameras installed.

Also, we’re still mad about New Orleans killing our secret pre-crime detection program back in 2018.

18. The ruthless exposure of the private lives of public figures drives far too much talent away from government service. The public arena—and the shallow and petty assaults against those who dare to do something other than enrich themselves—has become so unforgiving that the republic is left with a significant roster of ineffectual, empty vessels whose ambition one would forgive if there were any genuine belief structure lurking within.

Translation: The most corrupt people we’ve ever seen in government who stand to make us the most money ever are getting exposed for their on-the-job intoxication, shady deals, sexual harassment allegations, and outright lying. How dare you not give these ghouls grace when they keep buying our shit? Truly great men — and we do mean men — are beyond question. A random woman at a government agency 99 percent of Americans have never heard of, however, is fair game.

19. The caution in public life that we unwittingly encourage is corrosive. Those who say nothing wrong often say nothing much at all.

Translation: People are unfairly using the public communications platforms that we mine for mass surveillance purposes to complain about our open bloodthirstiness and crypto-fascism.

20. The pervasive intolerance of religious belief in certain circles must be resisted. The elite’s intolerance of religious belief is perhaps one of the most telling signs that its political project constitutes a less open intellectual movement than many within it would claim.

Translation: We’re sick of people saying our cofounder is weird for believing that regulating AI will spawn the antichrist, and if we mention the “War on Christmas” we’ll make more money. We’re undecided about whether any of this applies to tolerating the pope.

21. Some cultures have produced vital advances; others remain dysfunctional and regressive. All cultures are now equal. Criticism and value judgments are forbidden. Yet this new dogma glosses over the fact that certain cultures and indeed subcultures . . . have produced wonders. Others have proven middling, and worse, regressive and harmful.

Translation: Which cultures? Oh, you know the ones.

22. We must resist the shallow temptation of a vacant and hollow pluralism. We, in America and more broadly the West, have for the past half century resisted defining national cultures in the name of inclusivity. But inclusion into what?

Translation: Are you still with us after 21 points? Great. Welcome to the great mystery. It cost you way less to get here than joining Scientology. Here’s the final thesis: Immigration? Bad. Canceling billionaires? Bad. Giving us money to fight (((globalism)))? Good. Just hit us up on cashapp.

Apple’s long, bitter App Store antitrust war

2026-04-06T13:08:07-04:00

This is The Stepback, a weekly newsletter breaking down one essential story from the tech world. For more on the legal travails of Big Tech, follow Adi Robertson. The Stepback arrives in our subscribers’ inboxes at 8AM ET. Opt in for The Stepback here.

How it started

The year was 1998, and reigning personal computer giant Microsoft was on trial for violating antitrust laws, including by targeting its smaller competitor Apple. Apple occupied only a fraction of the PC market, while Microsoft held north of 80 percent. But its cross-platform QuickTime multimedia player threatened Microsoft’s own offerings, and a court determined that Microsoft had tried to crush it — pushing Apple to abandon a QuickTime version for Windows and implying it would limit the tool’s distribution options if Apple didn’t back off.

Anyone who’s used an electronic device lately probably knows Apple’s position has shifted. It may have never unseated Microsoft in the personal computer market, but it reigns in the far bigger category of mobile computing. It makes money at virtually every layer of its ubiquitous iPhone: the phone’s hardware, numerous accessories like earbuds and location trackers, first-party software services like Apple Music, and commissions from the developers whose apps populate the App Store. Even the iOS search bar is a moneymaker, thanks to a revenue-sharing deal that sets Google Search as the default.

All that power, combined with Apple’s tight control over its mobile ecosystem, has raised a lot of hackles. Some hardware and software developers say Apple copied and integrated tools they built (a practice known as Sherlocking), then disadvantaged them by locking them out of certain iOS features that its own tool could access — the former typically isn’t illegal, but the latter can be. Many app makers are critical of the App Store commission, pejoratively known as the “Apple Tax.” Developers and users alike are sometimes frustrated with Apple’s lack of support for third-party app stores or sideloading, which rival phone maker Google (albeit with its own anticompetitive restrictions) allows.

Over the past decade in particular, Apple has joined the growing number of major tech companies facing antitrust action. Chief among its critics is Fortnite maker Epic Games, which has filed legal complaints in several countries, seeking to use its own payment system and launch a third-party app store on iOS. Governments across the world — including in the US, the European Union, Brazil, Korea, and Japan — have also gotten in on the action, seeking to crack open the walls of Apple’s digital garden.

In an industry full of sprawling multipronged tech empires, the basic antitrust argument against Apple is comparatively simple: it’s become the ultimate gatekeeper to billions of people’s primary computing hardware, and it keeps competitors locked out while levying a heavy toll on the developers it lets through. The details are different, but in some ways, it hits the same emotional notes as the old case against Microsoft — they’re both stories about a company limiting what you can do with your personal device.

Navigating the legal implications of iOS’ design, though, has proven complicated. Actually changing it is proving even tougher.

How it’s going

Regulators and courts around the world have ordered changes at Apple, particularly around the App Store — but those changes have been slow to arrive, in part because for a half-decade or more, Apple has dragged its feet at every turn.

One of Apple’s highest-profile antitrust battles was the US lawsuit brought by Epic in 2020. Epic asked a judge to make Apple open up iOS to third-party app stores and alternate in-app payment methods. Apple mostly prevailed — in a 2021 ruling, Judge Yvonne Gonzalez Rogers largely accepted its argument that iOS’ walled-garden design provided real safety benefits and wasn’t unfairly anticompetitive.

But the company has spent years fighting over a comparatively small loss: an order to let developers add links or buttons to outside web-based payment systems. Courts have determined that Apple deliberately failed to comply with the order, including by adding a “prohibitive” fee to use it. (This wasn’t the first time it had tacked on this kind of fee, either — it failed to comply with Dutch regulators’ demands to allow third-party payments for dating apps in 2022, racking up tens of millions of dollars in fines.)

Apple also avoided becoming collateral damage in a different antitrust suit, US v. Google. That case found that Google had monopolized the search market through methods like its search deal with Apple. But a judge declined to ban such deals after Apple testified it could significantly damage its business.

In other countries, Apple has faced harsher demands — most prominently in the EU, whose Digital Markets Act (DMA) was designed specifically to create competition in the tech world. Under regulatory pressure in 2024, Apple started allowing third-party app stores on iOS in the EU. But it did so with a number of restrictions and additional fee structures that discouraged developers from switching over. A year later, it became one of the first companies (alongside Meta) to face fines for violating the DMA, with the EU citing “overly strict” requirements and the new fees. Beyond the App Store, Apple has also avoided bringing some device features to the EU, including Live Translation for AirPods and iPhone Mirroring; it’s blamed the difficulty of supporting these features on third-party devices per DMA rules.

Despite Apple’s steady opposition, there have been tangible changes. For over a decade, it was impossible to actually buy ebooks through Amazon’s Kindle iOS app, for instance — but in mid-2025, Amazon used the US court order to start including “Get Book” links. The alternative iOS app store AltStore has launched in the EU and Japan, with plans to expand to Brazil and other countries; Epic has launched its Epic Games Store on iOS in Europe too. While Epic hasn’t released numbers for iOS store popularity, AltStore said it had “hundreds of thousands of users” as of last October. And in China, Apple recently reduced developer fees in attempts to avoid a potential investigation.

But for many people, antitrust action hasn’t massively changed the iPhone experience. A different EU third-party store, Setapp, shut down earlier this year citing “still-evolving and complex business terms”; Apple and the EU are sparring about who’s at fault. iOS remains effectively one of two global smartphone platforms, and Apple retains tremendous power at every level of it.

What happens next

Apple will likely keep tangling with governments. More countries, like Australia, have pushed pro-competitive regulatory overhauls. In 2024, the US Department of Justice filed an iOS-related antitrust suit against Apple, and it’s slowly moving toward trial — though judges can be leery of ordering drastic remedies even if companies are declared monopolies. Meanwhile, Chinese regulators seem poised to keep pushing for more changes — which could become a pressing issue for Apple in the coming year.

The EU and Apple will also continue hammering out what DMA compliance looks like for iOS. Apple initially planned to roll out a new fee structure at the start of 2026, but it’s claimed the EU “refused to let us implement the very changes that they requested,” failing to respond to a compliance plan and using “political delay tactics.”

For now, there’s a more immediate, non-regulatory potential threat to Apple: the rise of generative AI. Companies like OpenAI want to build a new computing pipeline that could bypass the existing system of phones and app stores, including by introducing their own devices. Apple has made comparatively few inroads into AI, and it remains dependent on other companies as it attempts to overhaul Siri with it. In theory, that could put it in the position of an incumbent tech giant about to be undercut by new technology — roughly the position that ’90s Microsoft found itself in with the web.

But Apple has survived other attempts to unseat it, like Mark Zuckerberg’s failed multibillion-dollar metaverse push. Losing the AI race hasn’t yet put a dent in phone sales. Early attempts at AI-first phone alternatives have been lackluster, and nobody’s figured out what an AI app economy looks like yet. So the battles over Apple’s power likely won’t stop any time soon.

By the way

Apple’s competitor Google manages a more open phone ecosystem with Android, but particularly in the US, it’s got a worse antitrust track record — it lost a legal battle with Epic that now seems likely to end in a settlement, and it’s been declared a monopolist in the search and ad-tech markets as well.
Long before App Store competition became a major concern, Apple fought a whole different, arguably weirder, antitrust battle over ebook publishing — after a 2012 DOJ suit accused it of conspiring with major publishing houses to shake Amazon’s dominance in the market. The case ended with a $450 million settlement.
Apple was one of the major targets of a 2021 US congressional push for antitrust reform, with witnesses from companies like Tile and Spotify relating stories about its allegedly anticompetitive conduct. Predictably for Congress, said push failed.

Read this

The Ringer has a classic oral history of the original Big Tech antitrust battle, US v. Microsoft.
Sean Hollister wrote about the complicated reasons why Apple mostly won its legal battle with Epic, while Google lost on The Verge.
Antitrust cases are a great chance to get an inside look at how companies function, and Epic v. Apple did not disappoint.
Cory Doctorow argues Apple’s “curated computing” model undercuts the company’s pro-privacy decisions and other positive moves.

Meta’s legal defeat could be a victory for children, or a loss for everyone

2026-04-03T11:59:14-04:00

Is social media not just bad, but illegally bad? Should tech companies pay for making it that way? According to two US juries — and no shortage of outside commentary — the answer to both questions is “yes.”

Earlier this week, two juries — one in New Mexico, one in Los Angeles — held Meta liable for a total of hundreds of millions of dollars for harming minors. YouTube was also found liable in Los Angeles, and both companies are appealing their losses. In one sense, the decisions were surprising. Meta and Google operate platforms for transmitting speech and are typically protected in a variety of ways by Section 230 and the First Amendment; it’s unusual for suits to clear these hurdles. In another, it feels inevitable. The web of 2026 has become almost synonymous with a few widely disliked for-profit platforms, and the harm they’ve caused is often tangible — but it’s still far from certain what this defeat will change, and what the collateral damage could be.

If these decisions survive appeal — which isn’t certain — the direct outcome would be multimillion-dollar penalties. Depending on the outcome of several more “bellwether” cases in Los Angeles, a much larger group settlement could be reached down the road. Even at this early stage, it’s a victory for a legal theory that social media platforms should be treated like defective products — a strategy designed to get around the shield of Section 230, but one that’s often failed in court. “The California case specifically is the first time social media has ever had to face the staredown and judgment of a jury for specific personal injuries,” attorney Carrie Goldberg, who pushed forward major early social media liability suits, including an unsuccessful case against Grindr, told The Verge. “It’s the dawn of a new era.”

“It’s the dawn of a new era.”

For many activists, the overall goal is to make clear that lawsuits will keep piling up if companies don’t change their business practices. What practices? In New Mexico, a jury was swayed by arguments that Meta had made statements misleading users about the safety of its platforms. In LA, the plaintiffs successfully claimed Instagram and YouTube were designed in a way that facilitated social media addiction that harmed a teenage user. Meta and Google (and other nervous companies) could plausibly change specific features or be more cautious in their public statements and disclosures. But each case depends on a set of highly specific circumstances, and there’s no one-size-fits-all answer about what needs to change.

Eric Goldman, a legal blogger and expert on Section 230, sees clear legal danger ahead for social media services. “These rulings indicate that juries are willing to impose major liability on social media providers based on claims of social media addiction,” Goldman wrote after the ruling. In an email to The Verge, he noted the issue was bigger than just juries. “Judges are certainly aware of the controversies around social media,” Goldman said. In the Los Angeles case and other upcoming bellwether trials, “the judges have not given social media defendants much benefit of the doubt, which is how the plaintiffs’ novel cases were able to reach trials in the first place.” It’s a situation, he says, that “does feel differently compared to a decade ago.”

Goldman pointed out that New York and California have also passed laws banning “addictive” social media feeds for teens — so even if an appeals court reverses the recent decisions, that won’t necessarily turn back the clock.

The best-case outcome of all this has been laid out by people like Julie Angwin, who wrote in The New York Times that companies should be pushed to change “toxic” features like infinite scrolling, beauty filters that encourage body dysmorphia, and algorithms that prioritize “shocking and crude” content. The worst-case scenario falls along the lines of a piece from Mike Masnick at Techdirt, who argued the rulings spell disaster for smaller social networks that could be sued for letting users post and see First Amendment-protected speech under a vague standard of harm. He noted that the New Mexico case hinged partly on arguing that Meta had harmed kids by providing end-to-end encryption in private messaging, creating an incentive to discontinue a feature that protects users’ privacy — and indeed, Meta discontinued end-to-end encryption on Instagram earlier this month.

“Judges have not given social media defendants much benefit of the doubt.”

Blake Reid, a professor at Colorado Law, is more circumspect. “It’s hard right now to forecast what’s going to happen,” Reid told The Verge in an interview. On Bluesky, he noted that companies will likely look for “cold, calculated” ways to avoid legal liability with the minimum possible disruption, not fundamentally rethink their business models. “There are obviously harms here and it’s pretty important that the tort system clocked those harms” in the recent cases, he told The Verge. “It’s just that what comes in the wake of them is less clear to me.”

While Reid sees legal risks for smaller platforms with fewer resources in these decisions, he’s not convinced they’re more serious than the challenges new entrants already face in a hyper-consolidated online landscape built on massive amounts of data collection. “There are things that make it hard to do something really new in this space that are driven by the sort of marketplace and the surrounding policy,” he said.

Reid, Goldman, and Masnick all warn there’s a clear chance that the fallout could harm marginalized people who use social media to connect. “There will be even stronger pushes to restrict or ban children from social media,” Goldman told The Verge. “This hurts many subpopulations of minors, ranging from LGBTQ teens who will be isolated from communities that can help them navigate their identities to minors on the autism spectrum who can express themselves better online than they can in face-to-face conversations.”

If platforms like Instagram are inherently damaging and directly comparable to gambling or cigarettes, comparisons frequently made by critics, being kicked off would be no great loss. But even research that suggests social media can be harmful for adolescents has associated moderate use with better well-being. Conversely, harmful online content like harassment and eating disorder communities still flourished before recommendation-driven, hyper-optimized modern social media; tinkering with specific algorithmic formulas could have a positive impact, but it’s possible it won’t provide a deep or lasting fix. The appeal of punishing Meta is obvious — what it will mean for everyone else is much less clear.

Meta’s reckoning over kids safety is in the hands of two juries

2026-03-24T10:56:31-04:00

Mark Zuckerberg. | Image: The Verge | Photo: Bloomberg via Getty Images

Two juries are currently deliberating a series of cases that could either usher in a legal reckoning for Meta, or maintain the status quo in an uphill battle to impose changes or penalties on tech platforms in court.

Yesterday, a New Mexico jury heard closing arguments in a trial where Meta is accused of facilitating child predators on its platforms — allegations the company vehemently denies. And as soon as today, a Los Angeles jury is tentatively expected to reach a verdict in a separate case, which concerns whether Meta and Google should be held liable for making defective products that addicted a young woman. Verdicts against the company could result in damages and civil penalties that could exceed $2 billion dollars. Perhaps more significantly, such an outcome could also invite more legal action after years of failed or stalled attempts to sue tech companies over alleged harm.

It’s already just the tip of the iceberg for Meta, as well as many other tech platforms, that are set to face several more trials this year. Meta’s products, Facebook and Instagram, have often been at the forefront of criticism over the tech industry’s alleged failure to protect kids online, fueled by leaks from former employees like Frances Haugen. Meta, meanwhile, argues that harming users is not good for business.

“While New Mexico makes sensationalist, irrelevant and distracting arguments, we’re focused on demonstrating our longstanding commitment to supporting young people,” Meta spokesperson Andy Stone told The Verge in a prior statement. He also said the company “strongly disagree[s]” with allegations in the separate set of lawsuits playing out in California, and “are confident the evidence will show our longstanding commitment to supporting young people.” The jury in Los Angeles has been deliberating for just over a week, following a five-week-long trial.

During closing arguments in New Mexico on Monday, Linda Singer, an attorney representing the state, told the jury that Meta has failed to install adequate protections for young people on its services, and misled the public about the safety of its products. Throughout the six-week trial, the state presented evidence from Meta’s own internal discussions and state investigators’ undercover operations. “Meta chooses how to design its algorithm,” Singer said. “When you’re optimizing for a metric, the algorithm takes all of that data to get better. Right now, it’s getting better given that goal of showing engaging content. But Meta could choose to program its algorithm to get better at safety, to get better at integrity, to get better at things that keep kids safe.” While Meta has promoted numerous additional child safety features over the years, Singer compared them to “adding a filter to a cigarette. It doesn’t change the fundamental nature of the product or make it safe.”

“Meta could choose to program its algorithm to get better at safety.”

Both juries in New Mexico and California heard similar evidence — including testimony from a set of former Meta employees — about internal concerns over the platform’s guardrails, discussions about getting users onto Meta platforms young, and harms it was allegedly aware of but didn’t take sufficient action to address. Singer said Meta ignored clear signals of kids under 13 on its platform, even though it said they weren’t allowed on. One elementary school principal wrote to Instagram head Adam Mosseri that almost all her kids were on the app, she said.

New Mexico attorneys also presented evidence from their own law enforcement investigations that led to the arrest of three suspected child predators. Investigators used decoy accounts that claimed to be minors to lure suspects, and found they were flooded with new friend requests and sexual chats from adults, even when the decoy account repeatedly claimed to be a minor in messages. The state said three suspects’ accounts weren’t shut down until after New Mexico announced their arrests, even though Meta’s own systems had allegedly flagged policy violations repeatedly.

In the company’s own closing arguments, Meta attorney Kevin Huff argued that Meta had clearly disclosed the limits of its safety systems and taken action whenever possible, while the state had focused on a “small amount of bad content” and “cherry-picked” statements. “We believe the evidence has shown that Meta works incredibly hard to protect users including teens,” Huff said. He also argued that the state’s investigators used “hacked and stolen accounts” and real people’s images nonconsensually to lure predators, arguing they were “not trying to replicate a true teen experience.”

“We believe the evidence has shown that Meta works incredibly hard to protect users including teens.”

Singer disputed the claims. “I want to be as plain as I can possibly be on this point. This is not a hacked account, this is not an image of an actual adult. It’s an age-regressed image of Mr. Kitch,” Singer said, referring to a New Mexico investigator. Another image used in a decoy account, she said, was AI-generated. “After all of the evidence you’ve heard about the way that Meta put kids in harm’s way, after the fact that they failed to detect that his 13-year-old account is being chatted with by sex offenders, Meta had the audacity to question whether he placed someone in danger. When the scale of what Meta has done here is astonishing and absolutely contrary to what it has said.”

One key hurdle for plaintiffs in each of these cases is overcoming the fact that Meta is protected by Section 230 for liability over third-party content. Singer clarified early in her presentation that “when I say harmful content, I’m not talking about the nature of the content. I’m talking about Meta’s misrepresentations about what it knew about the harmful content that was present and recommended on its platforms.” Huff, conversely, drew the jury’s attention to Section 230 multiple times and said the state’s claim of misrepresentation “doesn’t even get out of the starting gate.”

Singer urged the jury to award the maximum amount in civil penalties if they decide that Meta willfully misled the public about safety and engaged in “unconscionable trade practices” under New Mexico law. If the jury agrees that all teen users in New Mexico were not properly informed of Meta’s risks and award the maximum of $5,000 apiece, that sum could total more than $2 billion.

Meta’s attorney, Huff, argued the state had presented “zero evidence” that teens were using Instagram because they weren’t informed of the risks and said the calculation of users under 18 was “based on a fake number that doesn’t represent the number of teens in the state.” (The state’s attorney said the count was drawn from Meta’s own numbers.) “There is no evidence that anyone ever saw any of the 42 misstatements” attributed to Meta among New Mexico’s teen user base, Huff argued — and therefore, no reason to grant a penalty for it at all.

America desperately needs new privacy laws

2026-03-23T11:53:06-04:00

This is The Stepback, a weekly newsletter breaking down one essential story from the tech world. For more on the dire state of tech regulation, follow Adi Robertson. The Stepback arrives in our subscribers’ inboxes at 8AM ET. Opt in for The Stepback here.

How it started

In 1973, long before the modern digital era, the US Department of Health, Education, and Welfare (HEW) published a report called “Records, Computers, and the Rights of Citizens.” Networked computers seemed “destined to become the principal medium for making, storing, and using records about people,” the report’s foreword began. These systems could be a “powerful management tool.” But with few legal safeguards, they could erode the basic human right to privacy — particularly “control by an individual over the uses made of information about him.”

These concerns weren’t just cheap talk in Washington. In 1974, Congress passed the Privacy Act, which set some of the first rules aimed at computerized records systems — limiting when government agencies could share information and outlining what access individuals should have. Over the course of the 20th century, the Privacy Act was joined by more privacy rules for fields including healthcare, websites for children, electronic communications, and even video cassette rentals. But over the past couple of decades, amid an explosion in digital surveillance by governments and private companies, Congress has repeatedly failed to keep up.

Lawmakers have weighed numerous plans for preserving Americans’ privacy, yet over and over, they’ve fizzled. Attempts to rein in government spying — like proposed updates to the Electronic Communications Privacy Act of 1986 — have been sandbagged by fears they’d compromise police and anti-terrorism operations. Despite multiple concerted attempts from members of both parties, Congress hasn’t passed a bill that governs how private companies collect data and what rights people have over their own information. Even highly targeted proposals like the Fourth Amendment Is Not for Sale Act — which restricts police from bypassing existing privacy laws by using data brokers — haven’t cleared the hurdle of becoming law.

Meanwhile, new technologies, from augmented reality glasses to generative artificial intelligence, create fresh risks every day — making it easier than ever to surreptitiously surveil people or encouraging sharing intimate information with tech platforms.

How it’s going

Immigration agents are harassing citizens that they’ve identified with data analytics tools and facial recognition. Data breaches at major tech companies are common, and security regulations meant to prevent them are being rolled back. Amazon just aired a Super Bowl ad bragging about how your doorbell can become part of a distributed surveillance dragnet for finding dogs.

At every point, invasions of privacy don’t just risk revealing something intimate about you to the world, they shift the balance of power toward whoever holds the most data. Take algorithmic pricing, where companies use personal information about shoppers to set individualized prices they estimate people will pay — resulting in companies like Instacart charging users different prices for the same item. (The company said this was an experiment it’s since ended.)

State-level and international regulations have addressed some privacy risks. Companies in Europe have been governed by the General Data Protection Regulation (GDPR) since 2018, though a rollback was proposed late last year. Several states have passed some form of general privacy framework, as well as more specific rules — Illinois’ biometric privacy law has facilitated lawsuits against Meta and others, for instance, and New York mandated algorithmic pricing disclosure a few months ago. However, privacy advocates warn many of the rules are inadequate. The Electronic Privacy Information Center (EPIC) and US PIRG Education Fund graded state consumer privacy bills in 2025, and only two states, California and Maryland, earned higher than a C.

EPIC deputy director Caitriona Fitzgerald tells The Verge that Congress has passed at least one meaningful reform lately: the 2024 Protecting Americans’ Data from Foreign Adversaries Act, which Fitzgerald calls “the strongest privacy law to be passed at the federal level in recent years.” PADFAA bars data brokers from letting hostile nations access sensitive personal information of Americans, and EPIC used it to file a complaint against Google’s real-time bidding ads system — which it alleges broadcast sensitive data indiscriminately.

Overall, though, it’s fair to say the situation isn’t great.

What happens next

As of early 2026, in many places, a sense of learned helplessness around privacy has taken hold. Companies like Meta push the line that if an existing technology already poses privacy concerns, it’s unreasonable to complain that a new technology does it even worse. According to internal documents, Meta also apparently believes that the Trump administration’s highly public flouting of civil liberties (or what Meta euphemistically deems a “dynamic political environment”) will keep activists distracted, leaving it free to push invasive features like facial recognition into products.

But the administration’s actions are making the dangers of these systems more and more difficult to ignore. It’s one thing to know the government could look up personal information about you. It’s another to have ICE agents intimidate you by dropping your name.

Not all of today’s privacy nightmares have easy regulatory solutions. But privacy groups have said for years that there are obvious ways to start improving the situation. A long-standing wishlist from a coalition that includes EPIC, PIRG, and others suggests creating a new independent federal Data Protection Agency, as well as a private right of action that would let individuals sue over violations of privacy laws. One of the most recent proposals is the Data Justice Act, a piece of model legislation outlined last month by a group of scholars at NYU Law. It’s aimed at limiting state collection and use of our deep digital footprints, aiming to redefine personal data “not as information the state may freely access, but as something inherently ours.”

There’s likely no turning back the clock on many digital technologies — nor, in many cases, would people want to. But it’s past time for more lawmakers to take the risks these technologies create seriously and decide it’s worth fighting back.

By the way

In many ways, governments across the world are actually going backward on privacy, thanks to the rise of online age-gating. In the US, the Supreme Court has already okayed age verification for sites with a large volume of adult content. Now, multiple states have passed laws that require it for essentially every app on your phone, a policy the Supreme Court seems likely to consider sometime this year.
Virtually every problem in tech regulation is intertwined, so tech monopolies also exacerbate privacy problems by reducing competition and concentrating information in a few places where it can be exploited. (That’s another issue Congress has taken up but failed to follow through on.) Also, laws don’t work if the government won’t fairly enforce them, so the Trump administration’s era of gangster tech regulation needs to end.
One of the simplest rallying cries for privacy in recent years is “ban facial recognition” — typically from use by government and law enforcement, but there’s a push to limit its rollout privately on smart glasses, too.

Read this

Smart glasses highlight just how narrow the reasonable expectation of privacy has gotten.
NYU Law’s Data Justice Act paper outlines a lot of specific problems that privacy reforms should tackle.
Julia Angwin writes about how Elon Musk’s Department of Government Efficiency weaponized government databases for surveillance.