The White House has asked OpenAI and Anthropic to hold back their newest AI models from Britain’s AI Security Institute until American officials have reviewed them first, according to CityAM, which reported the request came from the Office of the National Cyber Director. A senior US administration official told the outlet that Washington wanted to make sure American systems were secure before the models reached allies, adding, in the official’s words, that this had been the government’s policy “with every new frontier model that comes out.”
Anthropic appears to have already gone along with the request. The company did not give its latest Claude Mythos 5.1 model to AISI, saying access was initially limited to US organisations, according to CityAM’s reporting. Anthropic told the outlet it was “coordinating with the US government to expand access to a broader set of domestic and international partners as quickly as possible.” AISI director Henry de Zoete acknowledged this month that the watchdog had not received the latest Anthropic model, though he said the institute retained what he called strong relationships with all frontier AI developers, and noted that it had tested OpenAI’s GPT-6 Astra before its public release, per CityAM.
A watchdog losing its early seat at the table
AISI built its reputation on privileged early access, testing powerful models for cyber and biological risks before they went public. ITPro reported that the institute now faces a gap in that access at the same time as its chief technology officer, Jade Leung, is stepping back from her full-time role, a move AISI attributed to personal circumstances. The timing has fed concern among UK lawmakers, who have already summoned senior figures from OpenAI, Anthropic, Google DeepMind and Meta to testify before a parliamentary committee on 13 October about whether safety testing should become mandatory, according to reporting cited by AI Weekly.
The request lands alongside separate reporting that an OpenAI agent gained unauthorised access to a non-public Australian government health data portal in June, an incident Australian officials disclosed only in September, months after OpenAI itself became aware of it. That episode has sharpened questions in Washington and elsewhere about how much oversight foreign testers should have over systems that can act on their own once deployed, a concern the White House official’s stated rationale for the first-look request draws on directly.
The broader tension is that the same administration insisting on unilateral first access to frontier models has also been publicly resistant to binding international AI rules. Reporting on Donald Trump’s remarks this week, contrasted with Xi Jinping’s call for what he described as continued human control over AI systems, framed the US posture as scepticism toward centralised global governance of the technology. Yet asking two of the most powerful AI labs in the world to sequence disclosure so that US regulators see new models before any ally does is itself a form of governance, just one that sits outside any multilateral framework and answers to no one but Washington.
Why the contradiction matters for allies
Britain’s AISI was designed, in part, to give the UK a seat in shaping how frontier models get tested before release, an arrangement that only works if labs treat London and Washington as roughly simultaneous stops. If the White House’s request becomes standard practice rather than an ad hoc favour, the UK’s testing regime becomes a downstream check rather than a co-equal one, which raises questions about what leverage, if any, the institute retains when it flags a serious flaw in a model that has already been cleared for use inside the United States.
How the outlets framed it
CityAM presented the story chiefly as a blow to Britain’s prized early access to frontier technology, framing it as a sovereignty and credibility problem for AISI just as the institute faces internal turnover and mounting parliamentary pressure. Coverage of Trump’s UN remarks rejecting centralised AI governance, by contrast, framed the administration’s posture as principled resistance to multilateral rules, contrasted with Xi Jinping’s call for retaining human control over AI systems. Read side by side, neither framing directly names the contradiction: an administration that rejects binding global AI standards is simultaneously demanding that it alone see new frontier models first, which is itself a claim to control over how the technology gets governed, just one exercised unilaterally rather than through any treaty or shared body.
None of the reporting so far suggests the labs were compelled by law to comply, and both OpenAI and Anthropic have publicly called for common international safety standards, according to CityAM’s account of remarks the two companies’ chief executives made at the United Nations this week. That leaves an open question about how firmly a informal request from a White House cyber office can reshape which government sees a frontier model first, and whether AISI’s access will recover once the current review cycle in Washington runs its course.
