Telegraph: White House reverses course, UK AI Security Institute regains pre-release access to Anthropic, OpenAI and Google DeepMind models

Why it matters

【Key points】

The Telegraph reported on October 11 that the White House has reinstated the UK AI Security Institute's (AISI) pre-release access to frontier models from Anthropic, OpenAI and Google DeepMind. Last month the Office of the National Cyber Director asked these labs to stop sharing new models with AISI, and Anthropic then kept UK officials from reviewing Claude Mythos 5.1. The paper says the reversal followed lobbying by AI minister Kanishka Narayan, who is expected to say on Monday that tech firms cannot regulate themselves and to announce measures including changes to the Cyber Security and Resilience Bill requiring critical-infrastructure firms to report AI-related breaches.

【Verifiable】

The report is by Tony Diver (paywalled; syndicated in full on AOL). Politico first reported the September hold on September 24. The White House, AISI and the three labs have not publicly confirmed the reversal, or said whether the US-first testing order still applies.

【Our read】

1. What is at stake is the continuity of third-party testing. AISI is one of the few government bodies that publishes frontier-model findings; its GPT-6 Astra supply-chain finding this month is one example. If governments can switch that channel off, external evaluation becomes a diplomatic bargaining chip rather than an industry norm.

2. The terms are unclear. If the US keeps a test-first sequence, AISI may get later builds, which lowers the value of pre-release testing. Watch whether withheld models such as Mythos 5.1 are handed over.

3. The fix came through ministerial lobbying, not a formal agreement, so a similar cutoff could happen again.