(Image source from: AFP)
Anthropic announced on Friday that it will "quickly turn off" its most advanced AI models for everyone after the US government told them to stop foreign nationals from using these models due to national security worries. The company said it received the order to block access to Fable 5 and Mythos 5 for all foreign users, but did not get specific details on the national security issue. Anthropic mentioned they believe the government thinks there is a way to bypass a safety measure that would stop Fable 5 from being used to find software problems. This order comes as previous issues between officials from the Trump administration and Anthropic, which is planning for an IPO, were seeming to improve in various parts of the US government.
This year, things went sour between Anthropic and the government after the company declined to let the US military use its AI models for spying within the country and for completely automated weapons systems. In response, the government placed Anthropic on a blacklist, set to start later this year. This move represents a significant escalation in the US's efforts to prevent foreign rivals from enhancing their AI capabilities. For many years, US export rules have concentrated on the hardware and tools that make AI work rather than limiting foreign access to AI itself. According to Anthropic, the government has only provided "spoken evidence of a possible limited, specific jailbreak. " The company stated, "We do not believe that finding a narrow possible bypass should lead to pulling back a commercial model used by hundreds of millions of users. "
The government's order and Anthropic's reaction showcase the increasing conflict between AI developers and regulators about evaluating risks from so-called "jailbreaks," or techniques used to override model protections. Just a few days ago, Anthropic had requested more oversight of AI in the US, which would include the ability to prevent models that pose unacceptable risks. However, the company argued that the government's actions on Friday did not follow fair and evidence-based regulations. The Pentagon's chief information officer, Kirsten Davies, stated in a message on X that the Defense Department backed prioritizing national security. "Some matters are simply more important than income cycles, clickbait, and pre-IPO value. America First. Always," Davies expressed. Anthropic filed a confidential application for a US IPO last month, moving ahead of competitor OpenAI in the race to enter public markets.
Earlier this week, Anthropic introduced an AI model called Claude Fable 5, which represents a new level of abilities that they refer to as "Mythos-class. " This model comes with limitations preventing it from being used in dangerous fields like cybersecurity, which some users feel are too broad, according to Anthropic. Experts have indicated that if Mythos models are misused, they could significantly enhance advanced cyberattacks, especially in fields like banking that depend on intricate, connected, and often outdated technology systems. Anthropic mentioned that they collaborated with the US government and others on safety measures before launching Fable, and that rival AI models demonstrated a similar capacity to detect minor coding errors. "As a result of this directive, we have to quickly disable Fable 5 and Mythos 5 for all our users to stay compliant. Access to other Anthropic models will not change," Anthropic stated.
Anthropic expressed that they think there is a "misunderstanding" and are aiming to restore access to the models as quickly as they can. "If this rule was applied throughout the industry, we think it would essentially stop all new model launches for all leading model providers," the company commented.
Amazon's cloud service AWS announced late on Friday that Anthropic requested it to revoke access to the models for "all users in every region. " A US official confirmed that the Commerce Department had given an export control order to suspend access to Fable 5 and Mythos 5 for foreign users. Dean Ball, a former official in the White House who helped create the AI Action Plan issued in the summer of 2025, stated on X that the order indicates that all "non-Americans" would be prohibited from using Anthropic's latest models, including those located in the US.
"This means you should be ready to prove your citizenship to use Anthropic models," Ball remarked. Some important members of Anthropic, such as co-founder Chris Olah, AI researcher Andrej Karpathy, and philosopher Amanda Askell, were born outside the United States. Reuters could not find out their citizenship status, and an Anthropic representative chose not to comment on whether these individuals would lose access to the AI models.
