Anthropic skips UK safety review, stoking fears of Washington-only AI oversight

Anthropic's decision not to submit its latest model to a British AI safety institute comes as concerns are already growing about transparency in AI governance. The Trump administration has declined to make its AI security framework public, and prominent organizations from both the left and right have requested its release in a letter first shared with Semafor. That letter warns that “a lack of accountability over the governance of a [critical] technology would… likely lead to a loss of public trust.”

In that context, the UK's fear is not just about one model review but about where oversight happens. If frontier AI reviews become the sole province of Washington, British and other non-US safety bodies may lose the ability to examine the most advanced systems. Anthropic's Wednesday offer of “wide-ranging access” to an institute that conducted a review into cybersecurity incidents at OpenAI appears to be an attempt to keep some external scrutiny in place, though the body does not say whether that access would include the latest model or match the UK review's scope.

The stakes hinge on whether that alternative access is broad enough to satisfy safety reviewers and on whether the US framework is ultimately released. For UK officials and transparency advocates, the test is whether promised access substitutes for formal review, or whether it leaves a gap in accountability.