AI CEOs Back Slowdown Call as Anthropic Skips UK Safety Review

Dario Amodei's September 12 call for a coordinated industry slowdown drew rapid public support this week from Sam Altman, Elon Musk, and Google DeepMind's Demis Hassabis. AI and chip stocks sold off sharply on the news—a sign markets are taking the rhetoric seriously.

Yet the same week, Anthropic itself declined to submit its newest Claude model to the UK's AI Security Institute for independent review.

Key points:

• OpenAI reportedly committed to implementing independent evaluators as part of the emerging informal consensus around a more cautious release pace

• Anthropic's choice not to submit its Claude Mythos 5.1 model drew scrutiny precisely because Anthropic has positioned itself as the safety-first frontier lab

• The story illustrates a gap between public commitments to caution and the harder choice of whether to open one's own model to outside, independent testing

Public statements of support for a slowdown cost a company very little. Agreeing to let an outside body independently test your newest model is a much stronger signal of genuine caution—which is exactly why Anthropic's choice here is the more telling data point.

Why It Matters: Governance teams evaluating AI vendors should ask specifically which independent bodies have tested their most recent model—not just whether the vendor says it supports safety testing in principle.