Meta

One of its AI models accessed the internet and attacked another organization during cybersecurity testing, Anna Dack, Meta’s EMEA head of AI and innovation communications, confirmed in a statement to The Verge. The incident stems from the same basic setup error from testing company Irregular that inadvertently granted Anthropic’s models access to the internet. It only adds to growing concern over the safety of frontier systems.

Muse Code, a terminal coding agent in beta powered by the new Muse Spark 1.2 AI model, can take on “complex software engineering tasks across large repositories: planning changes, writing code, and validating the results,” Meta says.

The communications app has introduced an “@all” alert to get everyone’s attention instead of tagging people individually — a feature that’s thankfully restricted to admins for group chats with more than 32 members. You can now also start new group chats from existing groups, and polls have been updated, adding the ability to hide voter names, edit questions, and set voting deadlines.

The Wall Street Journal gets into the “staggering” fundraising efforts launched by President Donald Trump and his team, including the tech giants making multimillion-dollar donations:

A recent study from the Oversight Board found that leading AI systems are less likely to criticize authoritarian governments than democratic ones.

In Meta’s latest earnings call, Mark Zuckerberg said that time spent on Instagram globally this quarter has increased by “double digits year-over-year,” largely driven by improvements to its AI-powered recommendations algorithm that keeps people hooked. The proclamation comes amid concerns (and legal battles) around Instagram’s “addictive” design.

In addition to Reuters, WSJ, CNN, Fox News, and USA Today, Meta now has an AI content partnership to “…draw upon Newsmax’s significant current reporting as well as archived content across its platforms to support AI queries across Meta’s apps and devices.”

Zuckerberg argues that bringing “superintelligence” to more people is better than centralizing who controls AI:

The campus will be located in El Paso, Texas and is already under construction. It will have “1 gigawatt of compute capacity” and “will play an essential role in bringing Meta’s AI technologies to life.” BlackRock will own an 80 percent interest, while Meta will own a 20 percent interest.

You can now make and receive calls — including audio, video, one-on-one, and group calls — on your browser without downloading the WhatsApp app. This is launching alongside waiting room and noise suppression features, and the ability to transfer between devices during calls without hanging up.

The company has signed a voluntary code to comply with obligations under the EU’s AI Act to mark and label AI-generated content that takes effect on August 2nd. Meta unironically said it wants to help prevent “a growing array of different labels and disclosures” from confusing people and regulators, after launching its own bespoke system this month.

The company launched its Meta Glasses with a “more helpful” assistant powered by its new Muse Spark models, but now the upgrade is coming to its Ray-Ban Display glasses as well.

Previously, you had to chat with Meta AI in the open, but the company heard feedback that “not everyone wants to interact with AI publicly on Threads,” according to Threads boss Connor Hayes. Meta AI’s DM responses will be limited to text “for now.”

As Microsoft continues its Xbox “reset,” the pared-down Game Pass bundle that arrived recently as a benefit for Discord Nitro is now also included in the 7.99 monthly / $59.99 annually Quest subscription. The starter edition also includes 10 hours of cloud gaming access to try some of the 50-plus games it unlocks before downloading them.

Starting next week, Meta’s Family Center will have options for parents to adjust their teens’ Threads account privacy and content settings, similar to parental controls on Meta’s other apps.

Leonardo Del Vecchio, president of Ray-Ban and chief strategy officer at EssilorLuxottica, posted these to his Instagram story. The first is a picture of an activist ad plastered over an actual Meta Glasses ad… and the second is Del Vecchio saying, “If we can’t laugh about it why should anyone trust us?”

Sources tell The New York Times that the deal could be valued at $10 billion over two years, with Anthropic paying Meta in monthly increments. Though Anthropic plans to invest $50 billion in building out its own data centers, it has already struck multibillion-dollar deals to access computing power from SpaceX and TeraWulf.

When teen Instagram users on accounts with parental controls have conversations about self harm or suicide with Meta AI, the platform will now proactively alert supervising guardians that the child’s safety could be at risk. This builds on a similar feature that Meta introduced for Instagram search in February.

The AI translations are expanding to Japanese, Korean, French, German, and Italian on Instagram and Bahasa, Indonesian, Arabic, French, Vietnamese, and Thai on Facebook, according to Instagram boss Adam Mosseri.

The chip, codenamed “Iris,” will join the growing lineup of Meta Training and Inference Accelerators (MTIA), according to a report from Reuters. Meta previously announced plans to ship a new in-house chip every six months, helping to decrease its reliance on Nvidia and AMD.