After 17 days: US restores British early access to OpenAI and Anthropic models
After a pause of barely three weeks, the Trump administration has withdrawn its requirement that OpenAI and Anthropic deny British officials access to uncommercialized AI models. The arrangement is voluntary — and can be reversed as quickly as it was paused.
The United Kingdom can once again evaluate leading AI models before they are released. The country's AI minister, Kanishka Narayan, announced this on Monday, October 12, 2026, according to Bloomberg, which describes it as a reversal of a US policy that had required White House approval first (Bloomberg).
What Happened, and When
The timeline is tight: on September 24, 2026, the Trump administration asked OpenAI and Anthropic to stop sharing new models with the UK AI Security Institute (AISI), according to Crypto Briefing. On October 11, the administration reversed the directive and restored a voluntary arrangement that had thus been paused for only around 17 days (Crypto Briefing; Crypto Briefing).
After the September order, Anthropic had already denied British officials access to review its new model Claude Mythos 5.1, The Telegraph reports (The Telegraph via MSN).
Narayan summarized the British position to Bloomberg: the UK had worked closely with the US government "to retain access for Britain's AI security institute to the most powerful AI models before they are released" (Bloomberg).
What the Arrangement Is
AISI was established after the UK hosted the AI Safety Summit in 2023. The institute's task is to assess the capabilities of advanced AI systems and to search for potentially harmful uses. Pre-release access gives researchers the opportunity to test systems under controlled conditions before they reach the market (NewsBytes).
The arrangement is voluntary on the companies' side. The restored agreement gives AISI pre-release access to models from Anthropic, Google DeepMind and OpenAI — but no formal, written framework structure has been confirmed.
Details That Stand Out
One notable point: Google DeepMind was not named in the original September request, which was explicitly directed at OpenAI and Anthropic. Yet the restored access also covers DeepMind (Crypto Briefing).
It is also worth noting the context: when Trump spoke at the UN General Assembly in September, he said he wanted to "encourage" AI rather than "rein it in" (Firstpost). The September directive came in the middle of this line, and the October reversal does not change the underlying policy.
What We Do Not Know
Several matters remain open. It is unclear whether the US national security review that triggered the September directive was ever concluded — or what, if anything, it found. The sources provide no primary documentation: no official announcement from the White House, ONCD or AISI, only journalism that largely traces back to The Telegraph (hence an MSN-reproduced version with a somewhat unclear sourcing chain). That means we cannot say anything definitive about the legal or administrative form of the reversal.
It is also unclear whether Anthropic will now extend access to Claude Mythos 5.1 beyond US organizations, now that the arrangement has been restored.
Why It Matters
For the UK, this is about more than a symbolic victory. The country plans to use its G20 presidency next year to push for shared international principles and standards for advanced AI, and ministers are considering safety measures, including amendments to the Cyber Security and Resilience Bill (NewsBytes). Pre-release testing access is a concrete advantage in that work.
But this is no durable arrangement. The agreement is voluntary, based on goodwill, and can be reversed as quickly as it was paused. That is the central uncertainty in this story.