Anthropic CEO Dario Amodei called for a deliberate slowdown in the pace of AI capability growth in a roughly 4,000-word essay published on Saturday. In it, Anthropic commits immediately to giving external reviewers permanent access to internal systems – comparable to the access held by its own risk teams. OpenAI CEO Sam Altman and Tesla founder Elon Musk publicly backed the move the same day.
Amodei proposes a three-part plan for pacing
In his essay “We Must Pace the Frontier”, Amodei distinguishes pausing from pacing: Anthropic will neither stop training new models nor halt technical progress, but promises to make more time for safety work. His plan has three parts: external reviewers with permanent, employee-equivalent access, joint safety standards among leading companies under government mediation, and international coordination of development pace between democracies and authoritarian states.
Specifically, Amodei says external reviewers will get office access, company laptops, and rights comparable to those of internal risk teams. They are meant to verify safety practices, report incidents, and publish key findings on risk without editorial interference from Anthropic. Amodei cites a six-to-twelve-month window for urgent action, during which autonomous AI agent swarms could cause internet-scale damage – an assessment that comes from Anthropic itself and is not independently verified.
Altman and Musk back the move despite being rivals
OpenAI CEO Sam Altman responded on the platform X within hours, agreeing and announcing that his company would also deploy independent reviewers with employee-equivalent access. Pacing had already been a major topic of discussion inside OpenAI in recent weeks, Altman wrote, without giving further detail. Tesla founder Elon Musk, whose company SpaceXAI competes with Anthropic, briefly commented on the essay on X with the words: “Dario is right.” The backing thus spans three otherwise rival company chiefs – a rare show of unity at this level of clarity.
Amodei’s proposal builds on an open letter signed by more than 1,200 employees of leading AI companies in July 2026, which Amodei himself had co-signed. Unlike that appeal, which only asked the US government for tools, his current essay contains an immediate, unilateral commitment from Anthropic.
What will matter is whether the fast public agreement turns into binding rules. Amodei’s essay calls for government mediation on joint safety standards, but sets no timeline for negotiating them. Whether competitive pressure will actually force the three companies into restraint on their next model releases in the coming months will only become clear once a concrete frontier model is due.
