White House asks OpenAI and Anthropic to hold new models from UK testers until US review

The White House is requesting that OpenAI and Anthropic refrain from sharing their new AI models with the U.K. government's testing agency, the AI Security Institute (AISI), until these models have undergone testing by the U.S. government. This request comes from the Office of the National Cyber Director and puts the companies in a challenging position.

The AISI, a well-resourced government testing agency, has previously enjoyed privileged early access to models, but now may face upset from the Trump administration if they choose to withhold the models. Anthropic seems to have agreed to the request, stating that their Claude Mythos 5.1 model is only accessible to U.S. organizations.

OpenAI has not yet responded to the request. The White House's request aims to ensure that U.S. systems are secure before sharing models with U.S. partners. The U.S. government wants to review the models to ensure their security, while the U.K. has sought to maintain its close relationships with leading AI labs and emphasize the importance of global principles and standards to ensure AI development is safe.