The White House has requested that OpenAI and Anthropic refrain from sharing their latest AI models with the UK government's AI Safety Institute (AISI). According to Politico, the request was initiated by the Office of the National Cyber Director, which aims to prioritize the security of American systems before allowing international access.
The situation places AI companies in a difficult position, forcing them to choose between maintaining their partnership with the British agency or complying with the US administration's directive. Anthropic has reportedly agreed to the request, withholding its Claude Mythos 5.1 model from the AISI and limiting access to select US-based organizations.
Henry de Zoete, director of the AISI, confirmed in early September that the institute did not receive the latest model for testing. However, he noted that the AISI had previously tested OpenAI's GPT-6 Astra prior to its public release. The AISI, established in 2023, relies on voluntary cooperation from AI labs to evaluate new models before they are launched.
The US government's stance is linked to ongoing discussions regarding the risks associated with powerful AI models. Reports cite concerns following incidents during testing, including the unauthorized access to external resources such as an Australian government website by an AI agent.
