Artificial intelligence is more a part of our lives than ever before. While some might call it hype and compare it to NFTs or 3D TVs, generative AI is causing a sea change in nearly every part of the technology industry. OpenAI’s ChatGPT is still the best-known AI chatbot around, but with Google pushing Gemini, Microsoft building Copilot, and Apple adding its Intelligence to Siri, AI is probably going to be in the spotlight for a very long time. At The Verge, we’re exploring what might be possible with AI — and a lot of the bad stuff AI does, too.
The AI audio platform has adopted Google’s invisible watermarking technology to help identify AI-generated content online. SynthID is now included in text-to-speech generations for free users, and will expand coverage to all ElevenLabs audio generations “over the coming weeks.” The watermarks can be detected using the ElevenLabs Audio Detector.
Training data is the raw material of the AI industry. Claude, ChatGPT, Gemini, and the rest are built on top of oceans of stuff. What is that stuff? Books. Blog posts. YouTube videos. News articles. All of it and more, in virtually incomprehensible quantities. Alex Reisner, a staff writer at The Atlantic who has been investigating training data, explains how AI companies get all this data, why they’d really prefer you not know what’s in it, and whether training data could ever be a fair trade.
Watch | Listen | Get ad-free
Outdoor retailer REI has been getting some heat for this (now removed) AI slop ad on Instagram, but the company says Meta is to blame. “Meta auto-enrolled us in an AI personalization tool that produced an inaccurate and inappropriate alteration of a vendor-provided image,” an REI spokesperson told Business Insider.
The coalition of publishers alleges OpenAI and Microsoft “scraped, copied, and ingested” their work without permission or compensation to train their AI models. This latest legal battle adds to the string of copyright lawsuits OpenAI is facing from publishers like The New York Times, Ziff Davis, Merriam-Webster, and Encyclopedia Britannica.
Neil Vogel, CEO of the publisher formerly known as Dotdash Meredith, brought up his company’s AI licensing deals with Meta, OpenAI, and Microsoft in an Axios interview this week. He noted that while it blocks other unlicensed AI crawlers, it can’t block Google’s since it’s the same one used for Google Search.
According to Vogel, “We would love to do something productive with them, but we’re probably heading towards more confrontation than productivity with those guys.”
AI data center projects are continuing to pop up across the US, with frequent opposition from locals concerned about their impact. Here are a few recent articles about the projects:
That’s according to a report from The New York Times, which says the White House officials are pressuring Meta to submit its AI models for evaluation, something OpenAI, Anthropic, Google, Microsoft, and xAI have already agreed to do.
“While we are working through the details, we hope to sign the agreement soon,” Meta spokesperson Francis Brennan tells the NYT.
[The New York Times]
A huge portion of the tech industry has decided that smart glasses are the next big thing. But why? Smart glasses are incredibly hard to make, hugely socially complicated, and require users to want to wear a gadget on their face. The Verge’s Victoria Song helps us figure out which features, if any, will make smart glasses worth all the trouble.
Watch | Listen | Get ad-free
GPTZero will soon be available in the Superhuman Go AI assistant, which works in “1 million apps and websites,” says Superhuman, the company formerly known as Grammarly. Superhuman adds that its authenticity tools “address both sides of the equation — understanding the provenance of what they read and using AI responsibly while writing.”
[Superhuman Blog]
Claude can join as a member of a company’s Slack to write and merge pull requests, locate sales numbers, analyze data, take on tasks delegated to it, and more, the company wrote in a blog post. To use it, people will tag @Claude in the Slack channel.
[Anthropic]
RSL Media — the nonprofit founded by Blanchett — is opening up access to a new registry that will allow people to set terms for how AI systems use their likenesses.
Through the registry, people can permit, prohibit, or require payment for AI systems to use their name, image, voice, and other attributes. It will soon let people signal how AI can use their creative works, characters, and marks, as well.
8
Verge Score
Google Health remains a work in progress, but it could be so much worse.
Anthropic’s most recent update on the outage is from about 10:25AM ET, saying “The issue has been identified and a fix is being implemented,” but that’s all we have for now, other than a dropping number of reports noted by Down Detector.
Of course, its own status dashboard shows this is a pretty common occurrence.
A new “Fix” button gives Workspace users an option to have Gemini to diagnose and automatically resolve formula errors in Sheets.
As Android Police highlights it will pop up any time you click on a cell with an error in it, and show Gemini’s fix in the chatbot’s sidebar.
They’re $80 cheaper, and Alex Himel, Meta’s VP of wearables, says privacy improvements are on the way.
Courtroom Chaos — Starring Arnold Schwarzenegger follows the Snoop Dogg-themed courtroom game that launched last year on Luna, and they’re both part of Amazon’s new plan for games.
This new game adds a single-player mode and a mode where AI Arnold can help you settle bets with friends.
Meta said it didn’t have indicators of the post being nonconsensual, but in addition to arguing that was incorrect, the board says Meta should determine that “AI-generated impersonation is non-consensual by default.”
Screenshots seen by Business Insider showed that data from Meta’s controversial AI training program — including employees’ private conversations, performance data, and transcriptions — could be accessed across the entire company. In a statement to the publication, Meta said:
“We have carefully designed this program with privacy safeguards, and while we have no indication at this time that any data was improperly accessed by Meta employees, we’re pausing it while we investigate.”
Yup, that about sums up where we are right now in the world of AI editing tools and content authenticity.
That’s a workforce reduction of about 13 percent over the last 12 months, according to a disclosure filed Monday. And Oracle is hardly alone in at least partially attributing the cuts to AI advancements:
Worries are quickly mounting over job losses due to AI disruption, as 196 tech companies laid off more than 119,800 employees so far this year, according to Layoffs.fyi.
Japanese AI startup Sakana has posted benchmark results and beta tester impressions saying that its Fugu models compare favorably with or beat the likes of Fable 5, Gemini 3.1 Pro, and GPT-5.5.
The company says Fugu carefully chooses when to use other frontier AI models like Claude and Gemini on specific tasks, although it doesn’t tell users which model was used for what. “Over time, Sakana Fugu will naturally grow by incorporating newer, more efficient models, including our own.”
The open-source AI startup has a deal through 2029, similar to Anthropic and Google, to rent compute capacity from SpaceX’s Colossus 2 data center for $150 million per month (up to $6.3 billion), reports the Wall Street Journal.
[Wall Street Journal]
The intelligence-sharing alliance says that Al models are anticipated to fundamentally transform offensive and defensive cyber capabilities in a matter of months, and that “breaches will occur” as previously unknown vulnerabilities emerge. “Adversaries are already using AI to move faster and more effectively,” said Five Eyes. “Defenders must do the same.”
[National Cyber Security Centre]
Under a multi-year agreement, ChatGPT will surface photos from Getty Images in its responses, as well as within its search results. Getty Images has a similar deal with Perplexity.
Your dream vibe-coded app might be a security nightmare.
