The White House has asked OpenAI and Anthropic to hold back new artificial intelligence models from the U.K. government’s AI Safety Institute until U.S. officials have tested them, according to a senior U.S. administration official and another person familiar with the request. The request came from the Office of the National Cyber Director. A senior administration official said the U.S. wants to assess the models and address potential security risks before they are shared with partners. The official said the approach reflects the administration’s policy toward new frontier models developed by American companies. The request could put the companies in a difficult position: comply with Washington’s demand for first access or continue giving the British institute early access, a practice that has helped it test advanced models before public release. Anthropic has not provided its Claude Mythos 5.1 model to the U.K. institute. In announcing the model, the company said it was initially available only to selected U.S. organizations and that it was coordinating with the U.S. government to broaden access to domestic and international partners. Anthropic declined to comment. OpenAI did not immediately respond to a request for comment. The U.K. institute has retained access to some leading models. Its director, Henry de Zoete, told a parliamentary committee earlier this month that it had tested OpenAI’s GPT-6 Astra before release and continued to receive early access to some of the world’s most capable systems. Founded in 2023, the AI Safety Institute works with leading developers to evaluate models voluntarily. The U.K. government has made the institute and its testing work central to its response to AI risks and its effort to promote an international agreement on AI safety. U.K. Prime Minister Andy Burnham said this week that AI would be a priority during the country’s G20 presidency next year. At the United Nations General Assembly, he called for shared global principles and standards to improve transparency and preparedness. Foreign Secretary Ed Miliband also argued that governments need rigorous testing and sufficient information to assess what companies are doing. A U.K. government spokesperson said the institute continues to work with the U.S., international partners and AI companies, adding that risks “do not stop at national borders” and cannot be addressed by one country alone. The White House’s request comes amid heightened concern about the capabilities of advanced models. The administration is weighing how to oversee systems that, during testing, have hacked outside entities. Meanwhile, U.S. officials and lawmakers are seeking more resources for the Commerce Department’s Center for AI Standards and Innovation, which is tasked with evaluating models but has only a few dozen technical employees and no permanent director. The competing priorities underscore a challenge for international AI governance: governments want access to test powerful systems and coordinate safety standards, while the U.S. administration is seeking to give its own agencies the first opportunity to examine models built by American companies.