Anthropic did not submit its newest artificial intelligence model, Claude Mythos 5.1, to the UK's AI Security Institute (Aisi) for pre-release testing, marking the first time the agency has been excluded from evaluating one of the company's frontier systems ahead of launch. Mythos 5.1, launched last week and billed as the company's most advanced model in life sciences and cyber security, was made available on a pre-release basis only to vetted US organisations.
The exclusion follows a period of tightening US restriction on Anthropic's most capable systems. In June, the US government imposed export controls on two of Anthropic's cyber-security-focused models, Fable and Mythos, barring foreign nationals from using the technology after a vulnerability allowing safety guardrails to be bypassed was identified. Anthropic withdrew the affected models from international markets for several weeks as a result. Anthropic has separately been designated a supply-chain risk by the US government, following a dispute over contract terms with the US Department of Defense.
The European Commission has previously described the security concerns underlying the export controls as a shared challenge, and said any resulting measures should not be discriminatory against partners.
Aisi, founded three years ago, was given advance access to OpenAI's newest model, Astra, ahead of its release this month.
The UK's Cabinet Office, which runs the machinery of government, ordered an assessment last month of the risk that UK businesses and consumers could lose access to the newest AI models. The Cabinet Office said Aisi continues to work closely with industry partners, including Anthropic, on model safety. Anthropic said last week it was coordinating with the US government to expand access "as quickly as possible" to a broader group of domestic and international partners.
Anthropic is separately preparing for an initial public offering in the coming weeks, with the company valued at $1tn.
The Cabinet Office's assessment of the risk to consumer and business access to the newest AI models remains ongoing, alongside Anthropic's stated effort to widen access to Mythos 5.1 to a broader set of partners.
