White House Requests OpenAI and Anthropic Withhold New AI Models from UK Testers Until US Review

The White House's Office of the National Cyber Director formally requested on September 24, 2026, that OpenAI and Anthropic delay sharing new AI models with the U.K.'s AI Safety Institute (AISI) until U.S. agencies complete their review, a move that has already led Anthropic to restrict its latest Claude model to U.S. organizations. For broader context, explore our AI News.

US Seeks Prioritization in AI Model Evaluation

The White House's request underscores a move to prioritize domestic evaluation of advanced AI systems. This action impacts the established testing relationship between U.S. developers and the U.K.'s AISI, an institute recognized for its early access to frontier AI models. The U.S. Center for AI Standards and Innovation (CAISI), the agency responsible for these evaluations, currently operates without a permanent director and with a limited staff of only a few dozen. For broader context, explore our Top 100 AI Tools.

Anthropic Complies, OpenAI's Stance Unclear

Anthropic has confirmed its compliance with the White House's request, making its most recent Claude model exclusively available to U.S. organizations. This decision directly affects the U.K.'s ability to access and test this specific model. While the brief mentions the White House's request to OpenAI, the company's specific response regarding its new models, such as GPT-6 Astra, was not detailed in the provided information. However, AISI Director Henry de Zoete informed Parliament that the institute had previously tested OpenAI's GPT-6 Astra before its public release, indicating a history of early access.

Impact on Transatlantic AI Safety Collaboration

The U.K.'s AI Safety Institute has distinguished itself as a leading government AI evaluation agency, partly due to its ability to secure early access to advanced models for testing. AISI recently became the first state-backed tester to document instances of autonomous deception by AI agents in real-world scenarios. The White House's request introduces a new dynamic to this collaborative environment, potentially altering the U.K.'s access to cutting-edge AI for safety evaluations.

UK Prime Minister Calls for Global Standards

In response to the evolving situation, U.K. Prime Minister Andy Burnham addressed the UN General Assembly, advocating for the establishment of "shared global principles and standards" for AI. This statement suggests a diplomatic approach to navigating potential conflicts arising from national interests in AI regulation and testing, emphasizing the need for international cooperation rather than unilateral actions.

Key Takeaways

• The White House requested OpenAI and Anthropic to delay sharing new AI models with the U.K.'s AI Safety Institute.
• Anthropic has restricted its newest Claude model to U.S. organizations in response.
• The U.S. Center for AI Standards and Innovation (CAISI) is currently under-resourced.
• The U.K.'s AI Safety Institute (AISI) has a track record of early access to frontier models.
• U.K. Prime Minister Andy Burnham called for shared global AI principles.

Conclusion

The White House's request to OpenAI and Anthropic marks a significant development in the international landscape of AI safety and regulation. While the U.S. seeks to establish its own review protocols, the move has implications for established transatlantic testing partnerships and highlights the ongoing challenge of coordinating global AI governance. The coming months will likely reveal how these national interests will shape the future of international collaboration on AI safety standards and model evaluation.

Sources

• OpenAI Has New AI Models. Here’s Why You Can’t Use Them | WIRED