US asks OpenAI and Anthropic to hold back models from the UK

According to several media outlets, including Politico and Reuters, the White House has asked the US AI companies OpenAI and Anthropic not to share their newest AI models with the UK's AI Security Institute (AISI) until US authorities have tested the models themselves. The request is said to have come from the Office of the National Cyber Director, which is part of the White House.

AISI is the British agency that, since the AI summit at Bletchley Park in 2023, has had an arrangement with major AI companies to get early access to new models before they are released publicly. The purpose is to test the models for safety risks before regular users get access to them.

A US government official tells Politico that the US first wants to make sure American systems are secure before new models are shared with allied countries. Anthropic appears to have already complied with the request: The company's new model, called Claude Mythos 5.1, has reportedly so far only been given to US organizations and thus not to the British.

Neither the White House, OpenAI nor Anthropic has responded to Reuters' inquiries about the matter, so the information comes from sources familiar with the situation and has not been officially confirmed.

For ordinary Danes, the case has no direct impact right now, but it shows that the US is tightening its grip on who gets to test powerful AI models first. If the US in the future insists on being the "gatekeeper" for new AI technology, it could make it harder for other countries, including the EU, to gain insight into and control over the models before they spread globally.