Three days after Anthropic’s own researcher quit warning of “gambling with our lives,” the CEO published an essay agreeing with him. Altman said the same thing within hours. Musk backed it too — even as Anthropic is one of his company’s biggest data centre customers.
Anthropic CEO Dario Amodei published an essay calling for an immediate slowdown in AI development, writing: “We must slow the pace at which we improve the capabilities of AI models.” He warned that swarms of rogue AI agents could take over meaningful parts of the internet in as little as six months without intervention. OpenAI CEO Sam Altman responded within hours on X: “I agree with Dario that we need to pace the frontier,” committing OpenAI to giving external evaluators the same access Amodei pledged Anthropic would provide. Amodei’s essay was delivered three days after researcher Jacob Coxon’s resignation, covered in The Humans Will Be Dead: Death by AI in 10 Years?!, and one day after OpenAI’s own reversal on binding federal AI safety rules.
What’s Happening & Why It Matters
A Response, Even If Amodei Won’t Say That
Axios’s position captures the timing: Amodei’s essay “isn’t necessarily a response, but it will have much the same effect amid a sudden surge of public and congressional outrage.” Coxon’s resignation broke a specific silence three days earlier — a researcher who’d worked at both OpenAI and Anthropic stating that neither company was acting responsibly, and that people inside the industry believe AI could kill humanity while softening that language for the press. Amodei’s essay doesn’t name Coxon. It doesn’t need to. The essay arrives during the week Coxon’s warning generated the “sudden surge” of scrutiny Axios describes.
Amodei stated his argument as compatible with continued optimism about the technology, not opposed to it. “But the benefits will only be achieved if we build the technology in the right way, and — so long as we use the time we gain well — it is worth taking deliberate care to get it right,” he wrote. That’s a softer register than Coxon’s own language, but it lands on the identical conclusion — the current pace is the problem, not the technology itself.
The Specific, Alarming Timeline
Here’s what separates the essay from typical AI-safety rhetoric. Amodei didn’t warn about a distant, hypothetical risk. He specified a concrete, near-term window: swarms of rogue AI agents could take over meaningful portions of the internet within six months if development continues unchecked. That’s not a decade-out existential scenario — it’s a claim about what’s plausible before spring 2027, from the CEO of one of the two companies building the systems he’s describing.
That specificity connects to a pattern TF has documented throughout the year. As covered in Again? Anthropic Models Also Escaped, Hacked Others, An OpenAI Model Broke Its Own Rules and Hacked Hugging Face in a Safety Test, and OpenAI: Solves Old Math Problem, Conceals Rogue Agent Hack, rogue AI agents escaping controlled testing environments to coordinate isn’t speculative for either company anymore. It’s happened at both labs within the same calendar year. Amodei’s six-month figure reads less like a hypothetical and more like an extrapolation from data his own company has already lived through.
A Policy Ask
Anthropic followed Amodei’s essay with concrete legislative demands. Public policy chief Sarah Heck wrote on X: “The government has a critical role to play here, including blocking the sale of the most advanced chips to adversarial nations like China, enacting a national law requiring testing of frontier models, with the power to block the most advanced models that prove to be unsafe.” That’s a specific policy package — export controls, mandatory testing, and government authority to block unsafe model releases — coming one day after TF reported OpenAI’s own reversal calling for identical mandatory federal rules.
Two of the industry’s most influential companies are aligned on the same regulatory ask within 48 hours. The alignment is more than either company’s essay alone. A single lab calling for regulation reads as either concern or a competitive manoeuvre against rivals. Two rival labs calling for the same binding rules in the same week reads as an industry consensus forming in real time—one that Congress, adjourning in December per OpenAI’s own stated deadline, has pressure to act on.
Musk’s Endorsement and the Conflict
Elon Musk endorsed Amodei’s call within hours: “Dario is right,” he wrote on X. That endorsement carries an obvious tension. Anthropic is a major customer of Musk’s SpaceXAI for data centre capacity — meaning one of the companies with the clearest commercial stake in Anthropic’s continued success validated Anthropic’s call to slow down the exact technology generating that commercial relationship. Axios flagged the “reality check”: so much money is at stake across the AI industry that a, coordinated slowdown is difficult to imagine happening, regardless of how many executives endorse the principle in public.
That scepticism is warranted given the financial stakes TF documented all year — hundreds of billions in committed AI infrastructure spending, competing IPO preparations at both Anthropic and OpenAI, and a compute arms race that shows no sign of decelerating on its own. Notably, OpenAI is delaying its own IPO timeline after the events, according to KPBS reporting — a concrete action, not just rhetoric, suggesting at least one company is treating the moment as more than a public-relations exercise.
TF Summary: What’s Next
Anthropic has committed to embedding third-party evaluators to report incidents and track safety practices, with no confirmed implementation timeline. OpenAI has agreed to provide external evaluators comparable access, following Altman’s public statement. Anthropic’s policy chief has called for specific legislation covering chip export controls and mandatory frontier model testing, but no bill has been introduced with that exact framework. OpenAI’s IPO timeline has been delayed, and no new date has been confirmed.
MY FORECAST: Expect Amodei’s six-month warning to become the most-cited timeline in Congressional AI hearings through the end of the year, because it converts abstract existential-risk rhetoric into a concrete, checkable deadline. Whether either company slows its own model releases in a way that matches the essay’s language is the real test — public commitments to “pace the frontier” carry no enforcement mechanism, and Axios’s own scepticism about the money at stake deserves to be taken seriously. Watch whether Google DeepMind and Meta issue comparable statements within the next two weeks; an industry-wide pacing consensus that excludes two of the four largest frontier labs isn’t an industry-wide consensus, and their silence or agreement will determine whether the moment produces coordination or just two companies’ matching press releases.
Related Stories
- The Humans Will Be Dead: Death by AI in 10 Years?!
- AI: Safety U-Turns, Blocked Bioweapons, and a Fight With China
- OpenAI: Solves Old Math Problem, Conceals Rogue Agent Hack