The White House wants OpenAI and Anthropic to first provide new models for review by American agencies, and only then transfer them to the British Institute for AI Safety.
The available description does not specify which models are affected, which agencies will perform the review, or whether this is framed as a mandatory requirement. The package's sole source is The Decoder; its material is presented only as a brief metadata synopsis.
If this sequence becomes a working rule, it could add an American stage to international testing before models are handed to British experts. This is an interpretation, not a confirmed consequence.
editorial commentary
Why it matters
A likely consequence is an increased role for American oversight in international testing of models. The next observable signal will be an official confirmation from the White House, OpenAI, Anthropic, or the British AI Safety Institute. A significant uncertainty relates to whether the initiative has binding force and which models it covers.