White House Restores UK Access to Unreleased AI Systems
The Facts
- The White House has restored the U.K. AI Security Institute's access to unreleased AI systems from Anthropic, Google DeepMind and OpenAI after lobbying by AI minister Kanishka Narayan, allowing British officials to run safety tests before public release.
- Washington last month asked AI companies not to share their latest systems with the institute, with Anthropic subsequently barring U.K. officials from reviewing Claude Mythos 5.1.
- Both the U.K. AI Security Institute and the U.S. Center for AI Standards and Innovation were established following the 2023 Bletchley Summit, with a 2024 Memorandum of Understanding agreeing to "develop a shared approach to system evaluations, including the underpinning methodologies, infrastructures and processes."
Sources Split
LEFT
CENTER
RIGHT
+3
+1
+2
ESTAB
CRIT
CENTER
PRO
ESTAB
The Spin
Pro-government narrative
Independent, rigorous testing of frontier AI systems before release is the only thing standing between the public and systems nobody fully controls, making predeployment access for the world's best evaluators a necessary move. Building an independent evaluation muscle, diverse compute options and real legal teeth both allow friendly cooperation between allies and transform safety into a competitive edge in a changing world.
Government-critical narrative
Access granted by Washington can be revoked by Washington; a strategy built on borrowed American systems amounts to a castle built on sand. Renting cloud, chips and frontier access from a few U.S. giants leaves British defense one political mood swing from the dark, rendering sovereign AI systems, diversified compute and data kept on home soil the only real route to survival in a new technological era.
