On 12 September, Anthropic’s Dario Amodei published an essay calling for the AI industry to slow down the frontier, arguing “we owe it to humanity to try.” Within hours, Sam Altman posted “I agree with Dario that we need to pace the frontier,” Elon Musk replied “Dario is right,” and Google DeepMind’s Demis Hassabis nodded along. Four of the most powerful people in AI, who between them are spending hundreds of billions to build the thing as fast as possible, suddenly agreeing that maybe everyone should ease off. Forgive the raised eyebrow.
There are two ways to read this, and the honest answer is probably a bit of both. The charitable reading: these are serious people genuinely rattled by their own technology, and the resignation of Anthropic researcher Jacob Coxon (who quit warning the labs are racing toward something that could “kill us all”) clearly stung. The cynical reading: a “slowdown” that every major lab agrees to at the same moment is not a slowdown, it is a cartel with better PR, and it conveniently freezes the pecking order with the current leaders on top.
Watch what they do, not what they say
The tell is in the specifics, or the lack of them. Altman committed OpenAI to using independent evaluators with employee-like access, which is a real, checkable concession. He also announced OpenAI will not go public in 2026, saying safety concerns made it “an ill-advised moment” to IPO. That one is worth reading sceptically, because “we’re too responsible to take your money right now” is a remarkably flattering way to describe a company that may simply not want public-market scrutiny of its burn rate.
Musk saying “Dario is right” costs Musk nothing and lets him claim the moral high ground over rivals he is suing. Hassabis agreeing while DeepMind ships Gemini Robotics 2 the same week is its own kind of comedy. A slowdown everyone endorses and nobody quantifies is not a policy, it is a vibe.
The bit that is actually encouraging
To be fair, and this matters, it is genuinely better that the people building this are publicly nervous than smugly certain. The last time the industry spoke with one voice it was to insist regulation would kill innovation. A collective admission that the frontier is dangerous, even a self-serving one, shifts the ground under future policy fights, and independent evals with real access would be a concrete step if they materialise.
So hold both thoughts. The uncomfortable truth is that a voluntary, unmeasured, industry-wide “pause” endorsed by the very people it protects should be treated as marketing until proven otherwise. The genuine progress is that “move fast and break things” is no longer a sentence any of them will say out loud. We will believe the slowdown when we see a training run that does not happen. (Not investment advice.)
Related: the Anthropic researcher who quit over AI risk and the bill to ban superintelligence.