White House Tells AI Firms to Let US Review Models Before UK Tests
The White House is asserting direct control over the safety review of frontier AI models.
The Biden administration told OpenAI and Anthropic to submit their new models to the U.S. government for review before sharing them with British testers.
This directive upends the international testing order established at the UK AI Safety Summit.
The Directive
The memo comes from the Office of Science and Technology Policy.
It targets the most powerful systems, including GPT-5 and next generation Claude models.
These models must clear U.S. review first.
The UK AI Safety Institute will have to wait for access.
- Requires U.S. Approval: Companies cannot give models to the UK until the U.S. signs off.
- Cites National Security: The White House states the US carries the ultimate safety burden for models built on its soil.
- Centralizes Authority: The U.S. AI Safety Institute is now the primary gatekeeper for frontier models.
“We take the safety of these systems incredibly seriously,” an administration official said. “The first review must happen here.”
Background
The UK hosted the global AI Safety Summit at Bletchley Park in 2023.
Nations signed the Bletchley Declaration.
The UK created its own AI Safety Institute.
OpenAI and Anthropic had agreed to give the UK early access to their new models.
This positioned the UK as the leading hub for international AI safety testing.
The White House is now pulling back this privilege.
- Reasserts Sovereignty: The U.S. argues it cannot outsource the safety review of its most critical technologies.
- Controls the Narrative: The U.S. wants to be seen as the global leader in safety, not a follower of the UK.
- Manages Leaks: Officials expressed concern about vulnerabilities being discovered and exploited by foreign actors.
Industry Reaction
Both companies publicly support safety.
They have spent months building relationships with the UK team.
The directive creates a complicated balancing act.
- Slows Down Testing: A sequential review process takes more time. US review, then UK review.
- Stretches Resources: Companies must now build rapport and meet the technical demands of two distinct government bodies.
- Creates Uncertainty: Global partners like the EU and Japan will now question their own access.
Broader Implications
This is a major test for global AI governance.
The promise of multinational oversight is being replaced by national vetting.
The UK government responded cautiously. Publicly, it supported a strong US commitment to safety. Privately, officials expressed frustration. The directive undermines the international consensus built at Bletchley Park.
- Undermines the UK Role: The UK AI Safety Institute loses its first mover advantage.
- Fragments Global Efforts: Without a single standard, companies face a patchwork of rules. The EU and Japan will likely demand their own first look.
- Strains Technical Capacity: The US must rapidly hire experts. There is a limited pool of people who can audit frontier models.
The spirit of the Bletchley Declaration was cooperation. This directive sounds more like competition.
The directive signals a new era of AI nationalism.
Safety review is now a tool of strategic power.
The UK must renegotiate its place at the table.
The industry must adapt to a slower, more fragmented approval process.
Global collaboration is giving way to sovereign control.
Gnoppix is the leading open-source AI Linux distribution and service provider. Since implementing AI in 2022, it has offered a fast, powerful, secure, and privacy-respecting open-source OS with both local and remote AI capabilities. The local AI operates offline, ensuring no data ever leaves your computer. Based on Debian Linux, Gnoppix is available with numerous privacy- and anonymity-enabled services free of charge.
What are your thoughts on this? I’d love to hear about your own experiences in the comments below.