Anthropic and the U.S. Department of War — Photo: Reuters/Dado Ruvic/Illustration
Anthropic announced on Friday (12) that it has restricted access to two of its advanced artificial intelligence (AI) models. The company, which created the Claude family of language models, received an order from the U.S. government citing national security concerns.
In a statement, the firm said it had received a directive to block the models Claude Fable 5, launched last Tuesday, and Claude Mythos 5 for all foreign citizens, "both inside and outside the United States, including foreign employees of the company itself."
Access is temporarily blocked for all customers to ensure compliance with the order.
The sudden block marks a significant escalation in the standoff between Anthropic and the White House under President Donald Trump. Negotiations earlier this year over the use of the company's technology by the U.S. military and intelligence services failed.
The restriction could harm Anthropic's plans to conduct an initial public offering of shares, possibly in the second half of this year, with a valuation near US$1 trillion. Concerns among investors are growing over regulatory risks and the company's ability to maintain its technological edge.
Read also:
Anthropic suspends AI models after U.S. restricts foreign access
Anthropic or OpenAI: who will win the trillion-dollar dispute?
Risk of sophisticated cyberattacks
The blocked versions are the latest iterations of the Claude model. Fable 5 is widely available to the public, while Mythos 5 is largely restricted to pre-authorized organizations.
Fable 5 is a restricted version of Mythos 5, which Anthropic kept out of public reach due to concerns that it possesses "unprecedented capabilities" for identifying software vulnerabilities, some of which have remained unknown for decades, or code flaws that hackers could exploit.
This capability has so far been used by U.S. authorities and selected companies to patch security gaps. However, there has been fear from the start that this type of AI could become a dangerous cyber weapon in the wrong hands.
Experts say the Mythos models could significantly accelerate sophisticated attacks, especially in sectors like banking, which rely on complex, interconnected, and often decades-old technology systems.
Anthropic's own tests detected a small number of already known flaws, classified as "minor vulnerabilities," according to the company.
The firm refuted, however, that the removal of its products from circulation is justified, adding that if broadly applied, the rule "would essentially prevent the release of new models by all leading AI developers."
Rising conflict
The relationship between Anthropic and the U.S. government deteriorated this year after the company refused to allow its AI models to be used for domestic surveillance and fully autonomous weapons systems.
Claude is the most widely used cutting-edge AI model by the Pentagon and the only such model currently operating in Department of Defense systems that handle classified information.
In response, the Pentagon included Anthropic on a list of companies considered a risk to supply chains, which is set to take effect later this year and could heavily restrict its federal contracts.
Later, earlier this month, Trump signed an executive order requiring a prior assessment, for up to one month, of the most advanced AI systems for national security risks before their public release.
Until then, U.S. export controls had focused mainly on AI chips and hardware, rather than restricting foreign access to the models themselves.
Pentagon cites "national security"
Pentagon Chief Information Officer Kirsten Davies said in a post on X that the Department of Defense supports prioritizing national security. "Some things are simply more important than revenue cycles, clickbait, and pre-IPO valuations. America first. Always," she posted.
Anthropic confidentially filed for an initial public offering in the United States last month, moving ahead of rival OpenAI in the race to access public markets.
A few days ago, Anthropic CEO Dario Amodei voiced support for government blocking of potentially dangerous AI software. The company stresses, however, that this should occur based on transparent procedures, clear criteria, and technical facts—which, it says, is not happening at the moment.
The New York Times classified this week's order as "unusually broad," highlighting that it could prevent Anthropic employees in allied countries, such as Canada or the United Kingdom, from using the models.
Several key members of Anthropic, including co-founder Chris Olah, researcher Andrej Karpathy, and philosopher Amanda Askell, were born outside the United States. It is unclear whether they are U.S. citizens or whether they would be at risk of losing access to the AI models.