White House Asks OpenAI & Anthropic to Delay Sharing Models with UK Testers Until U.S. Review

The White House has asked OpenAI and Anthropic to hold off on sharing their newest AI models with UK testing agencies until the models have first gone through testing with the U.S. government, according to three people familiar with the matter.

The request — what the OpenAI Anthropic UK testers dynamic has come to represent — is a new front in the widening gap between U.S. and European approaches to AI safety. Britain has positioned itself as a hub for AI testing, setting up an AI Safety Institute that has begun evaluating frontier models. The U.S. government, by contrast, has so far resisted calls to create a comparable testing regime, even as it has secured commitments from major AI companies to share their models with U.S. officials before deployment.

The White House request is informal — it does not carry the force of law — but it reflects a broader U.S. effort to assert its authority over the testing of the most powerful AI models, even when those models are being evaluated by allied governments.

The people familiar with the matter said the request was made by the White House Office of Science and Technology Policy, which has been leading the U.S. government's engagement with the AI industry. The OSTP has been in regular contact with OpenAI and Anthropic in recent weeks, the people said, and has asked the companies to delay sharing their newest models with the UK AI Safety Institute until those models have been tested by U.S. government officials.

The UK AI Safety Institute, which was set up after last year's AI Safety Summit at Bletchley Park, has been in talks with OpenAI and Anthropic about testing their most advanced models. The institute has secured agreements to test models from both companies, and has begun evaluating them using a framework it developed with input from international experts.

The White House request, if followed, would delay those tests — potentially for weeks or months — while the U.S. government completes its own review.

The people familiar with the matter said it was not clear whether OpenAI or Anthropic had agreed to the request. Both companies have declined to comment.

The request highlights the tension between the U.S. and UK approaches to AI safety. The UK has positioned itself as a neutral hub for AI testing, bringing together experts from multiple countries to evaluate frontier models. The U.S., by contrast, has focused on securing voluntary commitments from AI companies — commitments that are not legally binding and that critics say are insufficient.

The White House request also comes amid broader tensions between the U.S. and UK over technology policy. The two countries have clashed over issues including semiconductor export controls, data privacy, and the regulation of social media platforms.

The people familiar with the matter said the White House request was not motivated by a desire to block the UK tests, but rather by a concern that the U.S. government should have the first look at the most advanced AI models before they are tested by other governments.

"It's about ensuring that the U.S. government has the information it needs to make informed decisions about AI safety," said one of the people, who asked not to be named because they were not authorized to discuss the matter publicly.

The request is the latest sign that the U.S. government is taking a more assertive role in the governance of AI, even as it has resisted calls for a comprehensive regulatory framework. In recent weeks, the White House has secured commitments from major AI companies to share their models with U.S. officials, and has begun working on an executive order that would establish a framework for AI safety testing.

The UK AI Safety Institute declined to comment. The White House did not respond to a request for comment.

Sources: Politico (September 24, 2026); Reuters (September 24, 2026); qz.com (September 25, 2026).

The bureau noted that AI would hurt less-educated workers most. That pattern is explored in AIPress's analysis of AI and working-class jobs — the groups with the least formal education, the lowest wages, and the least political influence are the ones most exposed to displacement, and the ones least able to absorb the shock when it comes.

For all the uncertainty, the direction of the trend is clear enough. Over the past three decades, the U.S. working class — particularly those without a college degree — has been losing ground on virtually every measure that matters: wages, employment, marriage, health, even life expectancy. AI threatens to accelerate that decline.