US reportedly restores British institute's access to unreleased AI models
The Telegraph reported on October 11 that the US administration had reversed course. The paper said the administration had asked developers not to share new models with Britain's AI Security Institute. Reuters says the reported change now lets American developers share models before release. The White House, OpenAI, and Anthropic did not immediately respond to Reuters.
Verified 12:21 AM PDT · 2 original sources
The reported change would let developers share models with the institute, but it would not require them to share anything. AISI says it does not certify models as safe or decide whether companies release them.
US officials or developers would need to confirm the change. The next question is whether AISI receives any models before release. Permission to share a model does not show that the institute tested it.
Audit the story
Original sources
Company claims remain company claims. Follow the reporting and judge the evidence directly.
- Reuters via Investing.comUS reverses course on Britain's testing of AI models, Telegraph says ↗
- UK AI Security InstituteOur approach to evaluations ↗
Continue the edition