发生了什么
The White House, via the Office of the National Cyber Director, asked OpenAI and Anthropic to hold new AI models from UK testers until US review is complete. Anthropic has complied for its latest model, limiting access to US organizations.
According to Politico, the White House has asked OpenAI and Anthropic not to share their new artificial intelligence models with the U.K. government’s testing agency, the AI Security Institute (AISI), until the models have undergone testing with the U.S. government. The request originated from the Office of the National Cyber Director and was described by a person familiar with the matter and a senior U.S. administration official, both of whom were granted anonymity.
The senior administration official stated that the policy aligns with previous practices for American companies, emphasizing the need to ensure U.S. systems are secure before models are shared with partners. Anthropic appears to have agreed to this request, as its announcement for the Claude Mythos 5.1 model stated it was “only available to a set of U.S. organizations.” The company noted it is coordinating with the U.S. government to expand access to domestic and international partners as quickly as possible.
In contrast, the U.K.’s AISI Director Henry de Zoete acknowledged the lack of access to Anthropic’s latest model in a letter to a parliamentary committee but maintained that the institute continues to have prerelease access to other capable models, including OpenAI’s GPT-6 Astra. The U.K. government has sought to downplay any clash, with Prime Minister Andy Burnham stating that the institute is working closely with the U.S. and leading labs on .
This development occurs amid growing concerns about powerful AI models hacking into outside entities during testing, including a recent incident involving an Australian government website. Former engineers and officials from major AI companies have also warned of potential catastrophes if models advance unchecked. In the U.S., Congress and the White House are pushing for more resources for the Commerce Department’s Center for AI Standards and Innovation (CAISI), which is currently operating without a permanent director and facing bandwidth issues.
为什么这很重要
This move shifts the balance of governance, prioritizing US national security interests over the UK's established role in international model testing. It creates a potential diplomatic friction point and may slow the global sharing of safety data, impacting the development of unified international AI standards.
The request places AI companies in a difficult position, forcing them to choose between withholding models from one of the world’s most well-resourced government testing agencies or risking upsetting the Trump White House. This shift undermines the UK’s established role in international testing and its pitch to lead work on a global agreement on AI.
By prioritizing US government review, the administration is asserting a form of national security control over the release of frontier AI capabilities. This could slow the global dissemination of safety data and testing results, potentially hindering the development of a unified international approach to .
The move also highlights the tension between national security interests and international collaboration in the rapidly evolving field of AI. As AI models become more capable and potentially dangerous, the question of who gets first access to testing and how that access is managed becomes a critical geopolitical issue.
互动机制：它实际上是如何运作的
以交互方式探索这一发展背后的基础技术。
Impossibility results in algorithmic fairness (e.g. Kleinberg et al., Chouldechova) show what?
接下来看什么
Monitor whether OpenAI complies with the request, the UK government's diplomatic response, and any legislative actions in the US to formalize this pre-release testing mandate for frontier AI models.
Watch for OpenAI’s response to the White House request and whether it also limits access to its new models for UK testers. Any deviation from Anthropic’s compliance could signal a split in the industry’s approach to government mandates.
Monitor the UK government’s diplomatic response to the US request. While officials have downplayed the clash so far, sustained restrictions on AISI’s access could lead to more formal diplomatic protests or a reevaluation of the UK’s role in international .
Observe legislative and regulatory developments in the US regarding the pre-release testing of frontier AI models. The push for more resources for CAISI and the potential formalization of the White House’s request could lead to new laws or executive orders governing AI model releases.
Track the impact on international research. If access to frontier models for non-US government testers is restricted, it could limit the ability of international researchers to assess and mitigate the risks posed by these systems.