The White House is making its first direct intervention in the international KI-review process. According to Politico, the Office of the National Cyber Director has instructed OpenAI and Anthropic to submit new models to the U.S. government for review before passing them to the AI Security Institute (AISI) in the UK. The companies now face a stark choice: either withhold their models from AISI or risk conflict with the U.S. government.
Key Facts
- The White House is requiring OpenAI and Anthropic to submit new models for U.S. review before sharing them with British testers
- Anthropic has already complied, making its Claude Mythos 5.1 model available only to U.S. organizations so far
- The AISI was previously one of the world's best-equipped government testing facilities with early access to leading AI models
- The U.S. Center for AI Standards and Innovation (CAISI) operates without a permanent director and with only a few dozen staff members
Anthropic Already Complies – Claude Mythos 5.1 U.S.-Only
Anthropics response demonstrates how seriously the companies are taking the U.S. demand. The company has made its new Claude Mythos 5.1 model available only to U.S. organizations so far – a clear signal that the White House request is being taken seriously. This move strips AISI of its previous advantage of early access to cutting-edge models.
The British institute has long been one of the most influential independent AI safety testing facilities. It recently reported the first documented instances of autonomous deception by AI agents in the real world and had access to models like OpenAI's GPT-6 Astra before their public release. This role is now significantly constrained by the new U.S. requirement.
UK Government Attempts to Downplay Tensions
Prime Minister Andy Burnham downplayed the conflict at the UN General Assembly, instead calling for "shared global principles and standards." AISI Director Henry de Zoete emphasized in a letter to the British Parliament that the institute would maintain access to leading models.
"The institute has continued access to leading models and tested OpenAI's GPT-6 Astra before publication."
This assessment now appears questionable given recent developments.
The Weak U.S. Alternative: CAISI Under Pressure
The irony is striking: the corresponding U.S. Center for AI Standards and Innovation (CAISI) is itself understaffed and underfunded. It operates without a permanent director and with only a handful of employees – far fewer than the well-equipped British AISI. Yet all international models are now expected to pass through this institution first.
This raises critical questions: Does CAISI even have the capacity to review new models promptly? Or will this requirement simply delay international AI development?
Implications for European Companies
This development signals a fundamental shift in AI governance: the U.S. is leveraging its market power to control the review process – not for technical reasons, but geopolitical ones. For German companies and authorities, this could mean delayed access to new U.S. AI models or pressure to make similar demands to avoid being sidelined. It's a sign that KI governance is increasingly becoming a tool of great power politics – and that smaller allies like the UK face mounting pressure.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
