Overview
- The US reversed a directive on October 11, 2026 that had paused the UK AI Security Institute’s access to pre-release systems from Anthropic, OpenAI and Google DeepMind.
- The hold began after the Trump administration asked OpenAI and Anthropic on September 24, 2026 to stop sharing new models pending a US national security review.
- During the 17-day pause, Anthropic made Claude Mythos 5.1 available only to US organisations, showing how labs changed distribution in response to the directive.
- UK AI minister Kanishka Narayan lobbied Washington for the reversal, which restored the earlier voluntary testing arrangement but did not disclose the review’s findings.
- The episode highlights that voluntary early-access testing is fragile, could push labs to alter releases, and leaves open whether governments will formalise cross-border safety checks; the AISI was created after the 2023 AI safety summit to probe such risks.