Britain is getting nervous that America’s AI tech bros are starting to pull up the drawbridge and keep their cleverest toys at home.
According to Financial Times Anthropic declined to give Britain’s AI Security Institute pre-release access to Claude Mythos 5.1, its latest and most capable life sciences and cyber security model.
For the first time, Anthropic restricted the model to vetted US organisations, leaving Aisi outside despite the agency becoming one of Britain’s few genuinely useful cards in the global AI race. The decision has rattled Whitehall, where officials increasingly worry that US AI outfits could fall in line with President Donald Trump’s more protectionist technology policies.
One government insider moaned: “National security officials have concerns. They definitely didn’t test Mythos 5.1. 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.”
Anthropic, valued at nearly $1 trillion and preparing for an initial public offering, launched Mythos 5.1 last week. The company bills it as its most advanced model for life sciences and cyber security. The uncomfortable bit for Britain is that access to frontier models increasingly looks like something the US can switch on and off according to domestic political priorities.
The Trump administration has taken a tougher line in recent months, temporarily restricting access to Anthropic’s most advanced models for foreign nationals, including employees working inside US AI companies. Some Whitehall figures suspect Anthropic withheld Mythos 5.1 from Aisi at the US government’s request. That suspicion remains unconfirmed.
Aisi received advance access to OpenAI’s Astra before its release this month, so the Americans have not yet welded the door completely shut.
Anthropic declined to comment on the British exclusion. Last week it said it was “co-ordinating with the US government to expand access to a broader set of domestic and international partners as quickly as possible”.
Translated from corporate soothing noises, foreign allies can apparently have the good stuff when Washington decides they are sufficiently deserving by helping to invade Iran or giving them Greenland.
The argument has been brewing since June, when the US government imposed export controls on Anthropic’s Fable and Mythos cybersecurity models and barred foreign nationals from using them. Those restrictions followed the discovery of a “jailbreak” that could bypass safety guardrails. Anthropic was forced to pull its cutting-edge models from international users for several weeks.
The episode annoyed Silicon Valley, which objected to the White House’s improvised approach to regulating frontier AI, but it went down even worse with supposed US allies. European governments complained that Washington was effectively punishing friendly countries by treating advanced AI capabilities as technology to be hoarded at home.
The European Commission warned that security threats were a “shared challenge” and said measures should “not be discriminatory against partners”. For Brussels, the problem goes beyond wounded diplomatic feelings. If the best US models can disappear whenever Washington changes its mind, European dependence on American AI starts looking less like an ecosystem and more like a vulnerability. Blighty has reached a similar uncomfortable conclusion.
Last month the Cabinet Office ordered an urgent assessment of the risk that ordinary users and businesses could lose access to the newest systems. It is not difficult to see why. Governments spent years accepting dependence on US cloud platforms, chips and software while Silicon Valley preached the wonders of an interconnected world.
Frontier AI now gives Washington another particularly useful tap to turn off. Anthropic’s relationship with the Trump administration has hardly been cosy. The company was designated a supply-chain risk after a dispute over contract terms with the US Department of Defence earlier this year. Relations appear to have improved, with Anthropic and the administration working closely on releasing Mythos and Fable.
That rapprochement will do little to cheer officials in London or Brussels if improved relations mean America’s AI firms increasingly regard foreign allies as second-tier customers.
