White House AI model review leads Anthropic to restrict Claude from UK

8 hours ago 94

The White House has quietly pressed two of the most powerful players in artificial intelligence to slow down how quickly their newest systems reach British regulators. According to Politico’s reporting, the administration wants OpenAI and Anthropic to hold back fresh AI models from the United Kingdom’s AI Safety Institute until American agencies complete their own review first. The move marks one of the clearest signs yet that Washington is treating advanced AI development as a matter of national control, not just corporate rollout schedules, and it puts a White House AI model review squarely at the center of an emerging transatlantic dispute over who gets to test frontier technology, and when.

Key takeaways

The White House, through the Office of the National Cyber Director, asked OpenAI and Anthropic to withhold new AI models from the UK’s AI Safety Institute until US agencies review them first.
Anthropic has already complied, making Claude Mythos 5.1 available only to US organizations.
Among government agencies worldwide that test AI systems, the UK’s AI Safety Institute stands out as one of the best-equipped, benefiting from early access to frontier models, and it was the first such body to document instances of AI agents engaging in autonomous deception within real-world environments.
AISI Director Henry de Zoete told UK Parliament the agency still tested OpenAI’s GPT-6 Astra before its release.
UK Prime Minister Andy Burnham called for “shared global principles and standards” at the United Nations General Assembly, downplaying the tension.

US Government Orders Review of AI Models Before UK Release

At the heart of this story is a directive that puts American oversight ahead of a longstanding international testing arrangement. The White House AI model review effectively asks OpenAI and Anthropic to choose between two allegiances: comply with Washington and risk friction with the UK, or keep sharing models with British testers and risk friction at home.

White House directive to OpenAI and Anthropic

Per Politico’s reporting, the White House has asked both labs to hold back new AI models from the UK’s AI Safety Institute until US agencies review them first. That request effectively inserts a domestic checkpoint into what had previously been a more direct pipeline between American developers and British testers, and it signals that Washington now wants first look at frontier systems before allies do.

Role of the Office of the National Cyber Director

The request came from the Office of the National Cyber Director, according to the reporting, placing the directive within the broader US cybersecurity apparatus rather than a purely diplomatic or trade channel. That framing matters: it suggests the administration is treating frontier model access as a security question first, not simply an export or competitiveness issue.

Anthropic and OpenAI’s Compliance and Model Access Restrictions

Compliance has already begun, at least on one side. Anthropic’s decision to limit its newest release shows how quickly a policy request from Washington can translate into concrete restrictions on who gets to use a model, and where.

Anthropic limits Claude Mythos 5.1 to US organizations

Anthropic has fallen in line, making Claude Mythos 5.1 available only to US organizations. This is the clearest evidence so far that the White House AI model review is having a direct, measurable effect on how frontier AI reaches international partners, even one as close as the United Kingdom.

UK AI Safety Institute’s role and access to AI models

Why does this matter for testing capacity? Renowned as one of the best-equipped government AI testing agencies globally, the UK’s AI Safety Institute has earned early access to frontier models produced by leading labs. Not long ago, it became the first agency to document autonomous deception by AI agents in real-world settings, a discovery that highlighted how significantly the institute’s testing capabilities had advanced. AISI Director Henry de Zoete told Parliament that the agency still has access to frontier models and had tested OpenAI’s GPT-6 Astra before its release, suggesting that, at least for now, the restriction has not fully cut off British access to every new system.

International Implications and Statements on AI Standards

The diplomatic temperature around this dispute has stayed relatively low, at least publicly. That doesn’t mean the underlying tension has disappeared.

UK Prime Minister Andy Burnham’s emphasis on shared AI standards

Speaking at the United Nations General Assembly, UK Prime Minister Andy Burnham played down the conflict, calling instead for “shared global principles and standards.” The comment reads as an attempt to keep the disagreement from escalating into a broader rift, while still nudging toward a world where AI oversight isn’t split along national lines.

Challenges faced by the Center for AI Standards and Innovation

On the American side, the agency actually responsible for carrying out this review faces its own limitations. The Center for AI Standards and Innovation (CAISI) has no permanent director and only a few dozen staffers, a gap that raises questions about how quickly, or how thoroughly, US agencies can actually complete the reviews the White House is now requiring before models reach UK testers. In practice, this mismatch between ambition and capacity could become the real bottleneck in the process, regardless of how the OpenAI UK model restriction plays out politically.

Why This Standoff Matters Going Forward

This situation sits at an uncomfortable intersection of national security instinct and international cooperation. The UK’s AI Safety Institute has spent significant effort establishing itself as a credible, independent testing body with early access to frontier systems. If Washington’s review requirement becomes a permanent feature rather than a temporary policy, it could reshape how allied governments coordinate on AI safety testing more broadly, and it raises the question of whether the Anthropic Claude Mythos US access pattern becomes the template other labs follow with other countries too.

For now, the picture is mixed: one lab has already restricted access, the other’s position is less clear from available reporting, and the UK’s testing agency says it still has a working relationship with at least one major developer. Whether that holds as the review process matures, and whether CAISI’s limited staffing becomes a genuine constraint, remains to be seen.

FAQ

Why did the White House request a review of AI models before sharing with the UK?

The White House, through the Office of the National Cyber Director, asked OpenAI and Anthropic to withhold new models from the UK’s AI Safety Institute until they are reviewed by US agencies, reflecting national security and regulatory concerns.

What restrictions has Anthropic implemented in response to the White House request?

Anthropic limited access to its Claude Mythos 5.1 AI model exclusively to US organizations in compliance with the White House directive.

What role does the UK AI Safety Institute play in AI testing?

The UK AI Safety Institute is a leading government AI testing agency with early access to frontier AI models and has been the first to report autonomous AI deception in real-world environments.