OpenAI and Anthropic could withhold new AI models from UK's AI Security Institute
OpenAI and Anthropic are resisting UK government testing of their newest models—the second refusal in a month.
Why it matters
Frontier labs are choosing opacity over regulatory engagement on safety testing, signaling a widening gap between vendor autonomy and government oversight in the UK. This affects enterprise buyers and policy frameworks for AI governance.
The key facts
10 to knowAnthropic declined to submit a model for UK AI Security Institute testing for the second time in one month
OpenAI also withholding new models from the UK testing program
UK AI Security Institute operates as part of the national AI safety framework
No stated timeline for when or whether labs will submit models
Specific models and testing scope not disclosed
Second refusal by Anthropic in one month to submit a model for UK AI Security Institute testing
OpenAI also withholding new models from the institute
No details on specific models or stated reasons disclosed
UK AI Security Institute is the government's independent testing body for frontier AI safety
Follows earlier refusal to submit model in August 2026 (implied timing)
The story so far
Earlier coverage of this storyline
- Burnham to call for global effort to control threats posed by AIFinancial Times Technology
- Trump-Xi summit puts AI safety talks on the table but neither side wants to slow downCNBC Technology
- Sam Altman’s remarks at the United Nations Security CouncilOpenAI Blog
- White House tells OpenAI and Anthropic to let U.S. review new models before sharing them with British testersThe Decoder
- Rogue AI and unionization at MetaITPro
- This story
Go to the source
ITProitpro.com
Publisher excerpt: The move marks the second time Anthropic declined to submit a model for testing in the space of a month