US Orders Anthropic to Cut Off Foreign Nationals From Its Most Advanced AI Models
The US government has forced an AI company to cut off its most advanced models from the entire world – including its own foreign employees.
On Friday at 5:21pm, Anthropic received a letter from the United States government. The letter did not explain its specific security concern in detail. It simply ordered the company to immediately suspend access to its two most powerful AI models – Fable 5 and Mythos 5 – for any foreign national, anywhere. Inside the United States. Outside it. Even foreign nationals employed by Anthropic itself.
Anthropic had no choice. It complied within hours. But because it could not reliably distinguish between foreign and domestic users on short notice, it had to disable the models for everyone. A capability that was available to users around the world on Thursday was gone on Friday.
The trigger, according to the company, was a report about what Mythos 5 can do. The model is particularly adept at detecting software vulnerabilities – including bugs that have remained undiscovered in critical systems for decades. This capability has been used by US authorities and selected corporate partners to plug security gaps. The concern, from the government’s perspective, is that the same capability could become a dangerous cyberweapon in the wrong hands.
The order was issued under US export control authorities, the same legal framework used to restrict the sale of advanced semiconductors and missile technology. This is the first time it has been applied to an AI model in this way – and the speed and breadth of the action suggest the administration is treating advanced AI as a strategic asset on par with nuclear technology.
Anthropic pushed back in its public statement, noting that it had reviewed the likely triggering report and concluded that the reported capability was narrow and specific: using the AI to review particular program code and correct errors. Models from other providers, including OpenAI’s GPT-5.5, possess similar capabilities, the company stressed. The implication was clear: if the government is singling out Anthropic, either the threat assessment is incomplete, or the policy is being applied unevenly.
Fable 5, which was released just this week, is built on Mythos technology but has its cybersecurity and biotech capabilities deliberately blocked. It was intended as a public-facing product. Mythos 5 itself is the non-public version, accessible only to government agencies and select corporate partners for system hardening. The order covers both.
The practical consequences are significant. Foreign researchers working at US universities who had access to these models have lost it. Foreign employees at Anthropic – including engineers who helped build the systems – are now barred from using them. The global AI research community, which already operates under restrictions on semiconductor exports, now faces a new wall: access to the most capable models is reserved for American citizens.
The long-term strategic question is whether this kind of restriction can work. AI models are not like fighter jets or microchips. They can be copied, downloaded, and run in places where US law does not reach. The open-source AI ecosystem already produces capable models that no government can control. By locking down its most advanced systems, the US may be ceding the global market to competitors who face no such restrictions – while gaining an uncertain security benefit at home.
Anthropic said it had received only partial information from the government about the threat that triggered the order. The company promised to seek more clarity and to work with authorities on a more targeted approach. For now, the models are dark – and the precedent has been set.