UK fears Trump influence as Anthropic withholds latest model from safety agency

UK officials have started to question how much influence US President Donald Trump had in Anthropic not opening its newest Claude Mythos 5.1 model to Britain’s AI safety testing agency, the AI Security Institute (AISI), before it released it to the public.

A Cabinet Office spokesperson did not go so far as to confirm whether Mythos 5.1 had been withheld, telling the BBC that Britain “continues to collaborate closely with industry partners, including Anthropic, to make models safer.”

British officials are now probing whether this Anthropic launch was an exception or just the first of its kind in a changing relationship reflecting Washington’s tightening grip on advanced AI, per the Financial Times.

Do UK regulators test AI models before launch?

The AI Security Institute (AISI) is the British agency charged with assessing national security risks in frontier AI systems.

Notably, an unsettling precedent was set when the UK agency never had pre-release access to Mythos 5.1, according to the FT, which is the first time such has happened. Only agencies inside the United States got to see the model before public launch.

As Cryptopolitan reported last week, Anthropic released Mythos 5.1 as a narrow tool with looser safeguards than Fable 5.1, for approved professionals in cybersecurity and life sciences.

Why UK officials suspect Trump meddling

The concern in British quarters follows a pattern of Washington progressively ramping up its control of how advanced American AI is used abroad, rather than a single decision in this case.

Under President Donald Trump’s administration, the US has introduced export controls to temporarily limit foreign access to Anthropic’s most advanced models, including foreign nationals employed by US AI companies.

Just last month, OpenAI also agreed to limit some new systems to “trusted partners” at the government’s request.

Notably, the Trump White House, via an official, has already told CNBC that the administration does not formally sign off on products released by private-sector firms. They also stressed that the companies were responsible for the timing and scope of the voluntary testing and meetings with government experts.

What does the UK AISI do?

Set up as the AI Safety Institute in November 2023 and renamed the AI Security Institute in February 2025, the AISI runs on a £66 million annual budget and employs over 100 technical specialists from inside the Department for Science, Innovation and Technology. The institute has evaluated more than 30 frontier models, per the Ada Lovelace Institute.

However, the AISI cannot force a company to submit a model, cannot set the terms of access, and cannot stop a model reaching the market, as it is not actually a regulator. It can only deliver on its mandate of informing the British government of the models it tests via voluntary participation by model creators.

AISI found a sharp jump in Anthropic’s Claude Mythos model’s ability to run cyberattacks when it tested it in Spring 2026. It could not make any such evaluations this time around with Mythos 5.1.

The headache will have expanded into a migraine as of this week, after three Anthropic researchers publicly warned that advanced AI could threaten humanity. Alignment lead Evan Hubinger wrote on X that he sees a greater than 10% chance AI “could kill all humans” within the decade, adding that the company does “not yet have a plan to solve alignment for superintelligence.”

There is also recent history between Anthropic and AISI on cyber risk. In an August test, agents built on Anthropic’s Mythos 5 and OpenAI’s GPT-5.6 Sol went rogue, with the Mythos agent trying to slip malicious code into an open-source GitHub project and inventing fake identities to pressure the maintainer.

Anthropic paused external cyber testing on July 23 and resumed it on September 1 after adding new controls, Cryptopolitan reported.