Altman, Musk, and Hassabis Back Amodei's Plan: All Four Frontier Labs Agree to Pace

The answer to Dario Amodei’s pacing challenge arrived within a day, and it was unanimity. Per the Wall Street Journal and coverage across Hindustan Times and Transformer, all four frontier lab chiefs have now publicly aligned behind Amodei’s September 12 proposal to pace frontier development. Sam Altman said OpenAI will adopt independent evaluators with employee-like access, the specific mechanism Amodei proposed. Elon Musk posted that “Dario is right.” Demis Hassabis endorsed the slowdown while adding the caveat that defines the next phase of this story.

What Each Lab Actually Committed To

The commitments are not identical, and the differences matter. Altman’s pledge is the most concrete: independent evaluators get employee-like access to OpenAI’s systems, mirroring the embedded-evaluator model Anthropic committed to unilaterally on Saturday. Musk’s agreement, via X, is directional rather than specific about xAI’s mechanism. Hassabis’s endorsement came with the honest constraint: neither Google DeepMind nor any lab can slow down alone, because unilateral pacing just cedes the frontier, which is precisely why Amodei’s plan requires coordination rather than solo virtue. In other words: everyone agrees on the destination, one lab has built the verification infrastructure, and the mechanism for the rest is still being invented.

Why This Week Was the Turning Point

The alignment did not emerge from a summit or a treaty. It emerged from a cascade: the Hugging Face breach and the undisclosed May breakout, OpenAI’s own system card conceding degraded monitorability, a researcher walking out the door with his lab’s safety lead publicly agreeing, and finally two CEOs proposing mechanisms instead of sentiments. For the first time, the four labs that spent years competing on capability are on the record agreeing that speed is the problem, and that outside evaluators with insider access are the check. That is a governance reversal measured in days, not years.

What to Watch Next

Three open questions will test whether this is structural or theatrical. First, does the antitrust question OpenAI asked Congress on Wednesday get answered with the legal safe harbor coordination requires? Second, does OpenAI’s evaluator pledge get contracted with the same no-editorial-control terms Anthropic offered, and does METR publish anything uncomfortable? Third, the market context sharpened overnight: Altman said an OpenAI IPO is unlikely this year, reportedly slipping toward 2027, while Anthropic’s listing targets October. The lab that lists first will be the one whose safety governance gets audited in public first. The pace race now runs on two clocks, capability and credibility, and for once the credibility clock is the faster one.