White House Requests OpenAI and Anthropic Not to Share AI Models with UK

The Biden administration has requested that OpenAI and Anthropic refrain from sharing their latest artificial intelligence models with the UK's testing agency. This request, reported by Politico, originates from the Office of the U.S. National Cyber Director.

The aim of the White House is to assess the safety of these models for American systems before allowing their deployment with international partners. This situation poses a challenge for the companies, as they must choose between withholding their models from the UK's Artificial Intelligence Safety Institute (AISI), which previously had early access, or risking potential conflict with U.S. authorities.

Reports indicate that Anthropic has complied with the request by withholding its Claude Mythos 5.1 model from AISI, limiting access to certain U.S. organizations only. Earlier, AISI Director Henry de Zute had informed a parliamentary committee that the institute had not received any new models for evaluation, although it did test OpenAI's GPT-6 Astra prior to its official release.

“The strong partnerships maintained by AISI allow us to test a range of models that demonstrate significant advancements across the industry,” de Zute remarked.

Formed in 2023, AISI routinely evaluates new models from prominent AI laboratories before their public launch. Politico also connected the White House's request to ongoing conversations about managing powerful AI models, particularly in light of prior testing phase incidents, including breaches involving external sites like that of the Australian government.

Additionally, on September 21, UN experts called for proactive strategies to prepare for potential loss of control over artificial intelligence systems.