The AI firm Anthropic did not provide the UK’s AI Security Institute (AISI) with access to Claude Mythos 5.1, its latest cyber-capable large language model, according to the Financial Times.
AISI is regarded as one of the world’s leading organisations for AI testing and subjects AI models to rigorous testing to assess how capable they are at offensive cybersecurity tasks. To date, it has tested all major model releases including Anthropic’s original release of Claude Mythos through trusted access prior to public launch.
But the FT reported that Anthropic has chosen not to provide AISI with access to Mythos 5.1, a model that it describes as its most capable for cybersecurity tasks and is proficient in potentially malicious activities. For example, Anthropic said in the wrong hands Mythos 5.1 could “significantly help” individuals or groups with basic technical backgrounds create and deploy chemical or biological weapons.
In the 24 hours following the announcement of OpenAI’s latest frontier model GPT-6 Astra, the minister for AI Kanishka Narayan confirmed that AISI had already tested the model. No such confirmation has been made for Mythos 5.1, which was announced on 1 September.
A government insider told the Financial Times that the move has left national security officials concerned.
“They definitely didn’t test Mythos 5.1,” the insider said. “There’s anxiety around it […] They are worried this could be a sign of things to come with AISI not being given access to the latest models.”
National security figures have increasingly warned the public over the threat of AI-enabled cyber attacks. In June, agencies from across the Five Eyes intelligence alliance warned that such attacks could come in months rather than years.
It is currently unclear whether the lack of access is an organisational policy, or if Anthropic prevented AISI from accessing Mythos 5.1 on the recommendation of the US government.
The Trump administration has imposed strict controls on Anthropic models throughout 2026. In June, the administration ordered Anthropic to block foreign nationals from accessing Fable 5 and Mythos 5, prompting the firm to disable them altogether.
The ban was subsequently lifted following negotiations with the government that saw Anthropic increase the safety margin used by Fable to block potentially harmful user requests.
In February, the Department of Defense designated Anthropic a supply chain risk following the firm’s refusal to provide it with a more permissive variant of Claude that could be used for domestic surveillance or autonomous weapons.
This classification was ruled unlawful in August, but US military officials have stated they continue to consider Anthropic a supply chain risk.
