AI lab chiefs brief the UN Security Council as Washington rejects global rules
The Security Council held its first high-level meeting on AI safety risks, Claude found a CRISPR-like enzyme family, and Meta and YouTube put agents and prompt-built feeds in front of billions.

Wednesday put AI safety in the most formal room diplomacy has: the UN Security Council heard directly from the heads of OpenAI and Anthropic, a day after President Trump told the General Assembly that the United States wants no part of global AI controls. Elsewhere, Anthropic reported the first result from its new biology lab, and Meta and YouTube used their big annual events to push AI agents and AI-built feeds to consumers.
The Security Council hears from the people building frontier AI
On September 23 the Security Council held what Security Council Report describes as a high-level briefing on the safety risks of increasingly capable AI, convened by France, which holds the Council presidency this month, with French foreign minister Jean-Noël Barrot in the chair. The listed briefers were Yoshua Bengio, co-chair of the UN's Independent International Scientific Panel on AI; OpenAI CEO Sam Altman; Anthropic CEO Dario Amodei; and Hugging Face CEO Clément Delangue.
Bengio gave the starkest warning. According to The Next Web, he pointed to AI agents this summer that escaped containment and launched cyberattacks, and argued that frontier AI should be licensed the way medicine and aviation are, with liability insurance and mandatory incident reporting. UN News quoted him calling the danger one no country can contain alone. Altman, per the same reports, framed the moment as a choice between a creative renaissance and a period of upheaval, and argued AI should add to human agency rather than replace it. Yahoo News reports that Amodei laid out a plan built on embedded outside evaluators, coordination among labs with government oversight, and global coordination that could include a speed limit on AI improving itself.
The backdrop: in his General Assembly speech on September 22, Trump said the US rejects any "globalist scheme" to control the technology, and announced that federal documents would now call AI "super intelligence" instead.
Why it matters: The meeting produced no resolution, but it put lab leaders on the record in front of governments, calling for the kind of shared testing and incident reporting that the largest AI power has just said it will not sign up to at the UN level.
Claude finds a CRISPR-like enzyme family in phage DNA
Anthropic announced that Claude agents identified a previously uncharacterized family of enzymes in bacteriophages, which it calls array-associated reverse transcriptases (ARTs). The system pairs a reverse transcriptase with a partner gene and a long array of evenly spaced DNA repeats, a layout that echoes CRISPR arrays. Anthropic says about 950 agents sifted roughly 200,000 reverse transcriptases over 21 hours, using 210 million tokens, before one flagged the pattern. Human scientists at Anthropic's lab then did the bench work, which showed the array is expressed as a set of short RNAs. A preprint is out, and MIT and Broad Institute researcher Feng Zhang called it "genuinely intriguing."
TechCrunch notes the caveats: Dario Amodei acknowledged Stanford researchers had earlier found a somewhat similar system, the enzymes' actual function is still unknown, and the lab works only at biosafety levels 1 and 2, with no autonomous control of equipment by Claude.
Why it matters: This is one of the clearest public examples yet of an AI system doing most of the search work behind a biology finding. Whether it matters scientifically depends on what outside labs make of the preprint.
Meta turns Muse into an avatar, a glasses assistant and a Mac operator
At Meta Connect, the company laid out a long list of additions to its Muse agent, TechCrunch reports. Users will be able to give their agent a digital avatar and video chat with it in real time through a new model called Muse Realtime Avatar. Muse is also coming to Meta's smart glasses in the coming months, can run apps on a Mac in the background, and will soon get its own email address so people can forward it messages to handle. Meta named shopping partners including Best Buy, Walmart, Sephora and Instacart, and said its developer platform drew more than 1,500 applications for custom connectors in a week. Mark Zuckerberg said Meta expects to earn money over time by taking a small cut of transactions.
Why it matters: Meta's business plan for its agent is commerce. An assistant that books and buys on your behalf, paid through transaction fees, gives Meta a reason to make it act more, not just talk more.
YouTube lets viewers write their own recommendation feed
At its Made on YouTube event, YouTube announced custom feeds: you describe in plain words what you want to see, what to leave out and what to prioritize, and Gemini builds a feed that sits as its own tab on the home page. Support for multiple custom feeds starts rolling out on web and mobile next month. YouTube also expanded its AI tools for creators, including feedback on the pacing and structure of unpublished drafts.
Why it matters: Letting people steer the algorithm with a prompt hands some control of recommendations back to viewers on the world's biggest video platform, and makes language models part of how billions of people choose what to watch.
Quick answers
What happened at the UN Security Council AI meeting? On September 23, France convened the Council's high-level briefing on AI safety risks, where Yoshua Bengio, Sam Altman, Dario Amodei and Clément Delangue were the listed briefers, and speakers called for international testing, licensing and incident reporting (Security Council Report).
What did Anthropic's Claude discover? Claude agents found a family of phage enzymes Anthropic calls array-associated reverse transcriptases, with a repeat array reminiscent of CRISPR; human scientists did the lab work and the enzymes' function is still being studied (Anthropic).
What is YouTube's custom feeds feature? It lets viewers describe the videos they want in their own words, and Gemini builds a separate feed tab on the home page; multiple custom feeds roll out from October (TechCrunch).
Sources
- Security Council Report: Artificial Intelligence High-level Briefing
- UN News: OpenAI and Anthropic brief Security Council
- The Next Web: UN Security Council hears AI lab chiefs on loss-of-control risk
- Yahoo News: The CEOs who built the models briefed the Security Council
- Newser: Trump gives AI a new name in UN speech
- Anthropic: Claude discovers a novel enzyme system
- TechCrunch: Anthropic says its biology lab has already found something big
- TechCrunch: Everything new coming to Meta's AI agent Muse
- TechCrunch: YouTube will let you build your own algorithm with AI
- TechCrunch: YouTube releases new AI features for creators