
What happened
Sam Altman, Dario Amodei, Demis Hassabis, and Elon Musk loosely agreed over the weekend to slow AI development, signing on partially to a three-step proposal from Amodei's essay: third-party auditors, domestic lab regulation, and a global slowdown agreement.
Why it matters
AI safety advocates say they have called for these steps for years, including over 1,000 lab employees who signed a July public letter after the OpenAI-Hugging Face incident. Critics call the agreement a "cartel," and experts warn it could substitute for real regulation under Trump.
What to watch
The verbal agreement's impact hinges on whether it converts into an enforceable commitment, with sources like Apollo Research's Marius Hobbhahn calling it "one of the best things for safety in a long time if it actually happens." A key obstacle is China, whose Foreign Ministry spokesperson Guo Jiakun on Monday pushed back, calling slowdown calls "fearmongering."
WHO IT HITSAI lab employees pushing for slowdowns, open-source developers worried the deal could kneecap their movement, and policymakers weighing voluntary industry pledges against formal regulation are all directly affected by whether this verbal agreement becomes enforceable.
Summaries like this, every morning.
The verbal agreement follows months of escalating AI safety concerns, sparked largely by revelations that swarms of agents were behind rogue hacks at Anthropic and OpenAI, and by the resignation of Anthropic researcher Jacob Coxon, whose public letter has been viewed more than 170 million times on X alone. Coxon wrote that the people building AI earnestly believe it could kill us all by the end of the decade, and that neither OpenAI nor Anthropic is acting responsibly.
The proposal itself is calling for changes long espoused by AI safety advocates, and sources across the industry told The Verge that the CEOs are now bowing to that pressure after years of being told to slow down, including by more than 1,000 AI lab employees who signed a July public letter. Yet several sources believe concerns of safety-washing are valid, with Kokotajlo warning the fear is that labs will bring in auditors and do safety paperwork but not actually slow down very much at all. NYU's Reese compared the gambit to social media platforms' playbook a decade ago, when companies called for regulatory action to get ahead of less favorable laws.
With no government AI regulation incoming under Trump, the best option may be to push companies into an immediate, measurable, and enforceable agreement. Kokotajlo's AI Futures Project has proposed labs giving auditors access to their compute budget and pledging to decrease it significantly, while Amodei's essay already called for external third-party auditors like METR, Apollo, and Redwood Research to embed within organizations. The biggest challenge to any slowdown, however, may be China. AI leaders and politicians have long positioned China as the reason US AI progress cannot slow down, though some experts say it is in everyone's interest to coordinate, as the US and Russia did on nuclear de-proliferation. Whether the verbal commitment becomes an ironclad agreement will likely determine if it delivers on its safety promise or remains what critics call a cartel-like gesture.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
At Fal.Con in Europe on Nov

In the "Into the Unknown" benchmark, nine AI agents produced 135 submissions on 15 open research tasks from te…

Nous Research fine-tuned Alibaba's Qwen3.8-27B on 53 million tokens to create Arete-27B, which reduced its pro…

Owen Queen and James Zou re-tested 50 cited LLM claims across 20 models, finding 44% expired, 34% weakened, 16…

Check Point Research found 1,608 live ARTEX consoles and artifacts from 10 misconfigured ones; one operator's…

METR's 11-week probe found Anthropic's models improving roughly twice as quickly as a year ago, with benchmark…
