Anthropic withheld latest AI model from UK testing agency

Reported: Anthropic did not give Britain’s AI Security Institute (Aisi) pre-release access to Claude Mythos 5.1 — billed as its most advanced life-sciences and cyber model — making it available initially only to vetted US organisations. First such exclusion of Aisi, which the piece calls a global leader in frontier-model testing; OpenAI’s Astra was shared with Aisi in advance. Whitehall alarm spans national security and economic risk if UK loses frontier access; Cabinet Office ordered an urgent assessment. Some officials suspect US administration pressure — the article stresses that suspicion remains unconfirmed. Anthropic declined to comment; said it is coordinating with Washington to broaden domestic and international partner access. Context: June US export controls on Anthropic’s Fable and Mythos after a jailbreak bypassed guardrails, briefly yanking models from global markets and drawing EU accusations of punishing allies. Anthropic (~$1tn, IPO weeks away) has had a tense US relationship (supply-chain-risk designation after a DoD dispute) though cooperation on Mythos/Fable improved. Cabinet Office says Aisi still collaborates with Anthropic and others on safety.