The US government has rescinded its earlier request that British inspectors not be given access to new AI models before their release. The news was reported by the Telegraph on October 11 — Reuters' coverage relies on the newspaper's report. Following the reversal, American companies such as OpenAI and Anthropic can once again share new models with the United Kingdom's AI Safety Institute.
According to Reuters' assessment, the move signals a softened US stance toward international AI safety initiatives: American developers gain broader freedom to cooperate with Britain.
How the Request Was Reversed
According to the Telegraph, the restriction imposed in September no longer applies. The White House has formally rescinded its earlier request not to share new models with the British institute — in other words, the US-imposed barrier that forced developers through an American safety review before handing models to foreign inspectors has been removed.
However, the information so far relies on a single source — the Telegraph report, which does not state the decision's exact date or who made it. Trump, OpenAI, and Anthropic representatives did not immediately respond to Reuters' requests for comment. Additional official details about the decision's exact form and its entry-into-force conditions are therefore awaited.
"Spokespeople for Trump, OpenAI and Anthropic did not immediately respond to Reuters' request for comment." — Reuters
It is unknown which model generations the decision applies to or whether it will cover future models — the Telegraph did not address those details. Nor has the White House yet explained why exactly it rescinded the earlier request.
How the Freeze Began
The earlier request emerged on September 24. According to Politico, the White House at that time asked OpenAI and Anthropic to withhold new models from British inspectors until US national security review was complete. The demand reflected the US intent to put its most advanced models through its own safety review first. As Politico stressed, it was not a complete ban but a temporary delay on sharing models with British inspectors until the American review concluded.
The restriction, in effect from September 24 to October 11, halted US–UK cooperation in international AI safety. The British institute had routinely tested the newest models of American companies before release.
What the British Institute Does
The United Kingdom's AI Safety Institute was established following the world's first AI safety summit, held in 2023. It tests the most advanced AI models, assessing their capabilities and misuse risk. The institute runs its tests before models are released publicly — allowing early assessment of both their new capabilities and their misuse risk.
The institute is a vital link in the international AI safety system: it conducts independent expert reviews before models launch and presents assessment results to policymakers. The earlier US request had blocked exactly this independent review path. Since then, the institute has exchanged testing experience with laboratories in the US, Europe, and Asia. As models grow ever more powerful, such advance testing matters more: policymakers gain the chance to learn about risks before mass release, not after.
Why Pre-Release Testing Matters
Pre-release testing is standard practice in the AI safety system. The institute tests models before public launch, evaluating new capabilities and hazardous-use scenarios. Such independent review differs from a developer's own testing: it provides an outside perspective and enables early vulnerability detection. According to Reuters' assessment, as models grow more powerful, policymakers have yet to find a definitive solution for overseeing them — independent pre-release testing is what strengthens that oversight system.
The US request of September 24 had halted exactly this practice. OpenAI and Anthropic — the two leading labs behind today's most advanced models — could not hand new models to British inspectors between September 24 and October 11. Now the barrier is lifted, meaning the institute returns to its testing schedule.
It is also worth noting that the freeze was tied to the US's own national security review: Washington intended to put new models through its own expert review first. The Telegraph report did not dwell on this.
What the Decision Means
According to Reuters' assessment, the rescission signals a softened US position toward international AI safety initiatives: American developers can work directly with British inspectors. At the same time, the US is tightening domestic oversight: Trump's newly established AI task force announced on Friday that developers must mandatorily report safety incidents in AI models — we wrote about this in detail in a separate article.
For now, these details rely on Telegraph and Reuters coverage; since the decision text has not been made public, the picture will become clearer once official documents appear. This article will be updated when official details emerge. Aishunos is following the story.
