Overview
- The White House reversed a September directive on October 11, 2026, restoring voluntary pre-release access for the UK’s AI Security Institute to models from OpenAI, Anthropic and Google DeepMind.
- The original U.S. request, issued on September 24, 2026, asked OpenAI and Anthropic to pause sharing new models with AISI until a U.S. national security review was completed, creating a temporary U.S.-first sequencing rule.
- Anthropic limited its Claude Mythos 5.1 release to U.S. organizations during the 17-day pause, showing how a policy shift can immediately alter product availability and rollout plans.
- UK AI Minister Kanishka Narayan lobbied for the reversal and is credited in reporting with helping restore AISI access, but officials have not disclosed whether the U.S. review is finished or what it found.
- The restored regime is voluntary and fragile, so stakeholders are watching for possible formal rules such as mandatory incident reporting that would reduce reliance on political decisions and developer goodwill.