Dario Amodei spent last weekend doing something no head of a frontier AI lab has done this directly: publicly asking the entire industry, including the company he leads, to slow down.
In a roughly 3,800-word essay titled "We Must Pace the Frontier," published on his personal website on Saturday, September 12, Amodei laid out an argument that AI capabilities are advancing faster than the industry's ability to verify they're safe — and that gap needs to be deliberately narrowed rather than closed after something goes wrong.
"We must slow the pace at which we improve the capabilities of AI models," Amodei wrote. "Progress will still seem fast, and we must make wise use of the time we gain."
A three-part plan, not a stop sign
Importantly, Amodei's proposal isn't a call to halt AI development. It's a call to pace it. The essay lays out three concrete steps. First, frontier labs would give embedded, employee-level access to independent third-party evaluators — drawn from groups like METR — with permission to test models and training processes directly rather than through limited external audits. Anthropic has already committed to this step unilaterally for its own models. Second, Amodei calls for coordinated safety standards among AI labs based in democratic countries, so that no single company feels commercially penalized for slowing down while competitors race ahead. Third, and furthest out, he floats eventual global coordination that would include China — acknowledging that a purely Western pact leaves an obvious gap.
Amodei explicitly frames this as pacing, not freezing: model training continues, but the rate at which new capabilities are unlocked would be intentionally throttled while safety verification catches up.
Rivals said yes — in public
What turned this from one CEO's essay into an industry moment was the response. OpenAI's Sam Altman posted that he agreed the industry "need[s] to pace the frontier," and said OpenAI would also expand access for external evaluators. Elon Musk, whose xAI competes directly with Anthropic, endorsed the call on X, writing "Dario is right" — notable given that Anthropic is also a major customer of Musk's data center capacity, so the two have a commercial relationship layered on top of the rivalry. Google DeepMind's Demis Hassabis was also reported among those who agreed.
Public agreement among leaders who are simultaneously racing each other for enterprise contracts, coding-agent market share, and frontier benchmarks is rare enough that it became the story in its own right, separate from the substance of Amodei's proposal.
Timing and context
The essay didn't arrive in a vacuum. Days earlier, a former Anthropic researcher resigned publicly and warned about AI risk, a moment that helped push AI safety concerns further into mainstream conversation and congressional attention. Whether or not Amodei's essay was a direct response, commentators noted it landed with far more force because of that backdrop.
The market noticed too. On Monday, September 14, chip stocks sold off in a move concentrated specifically in semiconductor names rather than the broader Nasdaq: Nvidia fell roughly 3%, AMD roughly 6%, and Intel roughly 7%. Investors appeared to be pricing in the possibility that a genuine industry-wide slowdown in capability gains could soften near-term compute demand.
Skepticism alongside the agreement
Not everyone read the essay as pure safety advocacy. Critics have pointed out that Anthropic has previously gone back and forth on the value of unilaterally slowing its own development, and some suggested the timing and framing conveniently double as a case for concentrating advanced AI development among a smaller number of well-resourced, safety-focused labs — Anthropic included — rather than a broader, more open ecosystem.
As of the essay's release, no competing lab had matched Anthropic's specific embedded-evaluator commitment with a formal policy of its own — the agreement so far has been rhetorical, not contractual. Whether that changes in the coming weeks will likely determine whether "Pace the Frontier" becomes an actual industry standard or remains a well-publicized essay with vocal supporters.
