Saturday’s unusual alignment is still the AI story on Sunday. Anthropic CEO Dario Amodei published a three-step plan to slow how fast frontier labs raise model capability. Hours later Sam Altman wrote that he agreed and that OpenAI would install independent evaluators with employee-like access. Elon Musk replied that Dario is right.
Amodei’s essay, circulated as “We Must Pace the Frontier,” argues safety work is no longer keeping up with capability gains since summer. He wrote that labs “must slow the pace at which we improve the capabilities of AI models,” and that progress would still look fast. The plan pairs outside scrutiny inside companies, shared industry standards, and cooperation among democratic governments without, in his telling, handing the lead to China.
Key takeaways
- Amodei: three-step slowdown — independent checks, common standards, international coordination.
- Altman: “pace the frontier,” plus a pledge of independent evaluators with employee-like access.
- Musk publicly backed the same line. No lab published a new training freeze timetable on Sunday.
What was actually committed
The concrete item is access, not a date. Altman said independent evaluators with staff-level access is a good idea and that OpenAI will do the same. Amodei had already pointed at METR-style outside review after Anthropic’s own cyber-eval disclosures this week. Musk’s agreement is a sentence, not a filing.
Whether any lab delays a named model is not in the Sunday record. The record is three rival CEOs saying the same clause in public.