ByREUTERS, TOBIAS HOLCMAN

JUNE 13, 2026 15:55

Anthropic announced on Friday that it would immediately restrict its leading artificial intelligence systems from all users following a US government directive to halt foreign access due to national security reasons.

The company was instructed to block access to Fable 5 and Mythos 5 for anyone outside the United States, yet officials provided no concrete explanation for the security worries, according to an Anthropic statement.

Anthropic understands the government is concerned about a possible workaround that could let Fable 5 detect software weaknesses despite built-in protections, the firm noted.

This directive arrives as earlier friction between the Trump administration and Anthropic, which is preparing for an IPO, appeared to be resolving in some federal agencies.

Tensions rose earlier this year when Anthropic declined to let the US military employ its AI tools for domestic monitoring or fully independent weapons. Officials then added the company to a supply chain blacklist scheduled to begin later this year.

The move represents a sharp increase in American efforts to limit adversaries' AI progress. Previous export rules targeted hardware and software tools rather than direct limits on AI model usage by foreigners.

Anthropic noted that officials shared only spoken indications of a limited, non-widespread workaround method.

The company stated it does not believe a narrow potential issue justifies withdrawing a commercial model already used by hundreds of millions of people.

The government order and Anthropic's reply underscore rising friction between AI firms and regulators regarding evaluation of so-called workarounds that bypass safety features.

Just days earlier, Anthropic had urged stronger US supervision of AI, including powers to stop models posing serious risks. However, the firm argued Friday's action did not reflect fair, evidence-based oversight.

Pentagon chief information officer Kirsten Davies posted on X that the Defense Department backs placing national security first.

Some priorities outweigh business timelines, sensational headlines, and pre-IPO values, Davies added, emphasizing America First principles.

Anthropic filed confidentially for a US IPO last month, moving ahead of competitor OpenAI in the push toward public listing.

Mythos Drives US Push for AI Development Rules

May coverage from major outlets highlighted how Anthropic's Mythos system is prompting Washington to consider tighter controls on AI creation.

Reports indicated Vice President JD Vance expressed concern after discussions with leading AI executives, noting Mythos raised alarms due to its capacity to independently spot software flaws.

The key issue is that such models might target vital local infrastructure managed by regional authorities lacking resources to counter attacks quickly, sources said.

National Economic Council Director Kevin Hassett stated the administration is developing regulations for new AI releases, modeled partly on FDA drug approval processes.

This approach would ensure models reach the public only after safety verification, though specifics remain under discussion, an official noted.

Advanced Cyber Threats

Earlier this week Anthropic introduced Claude Fable 5 as part of a new high-capability tier called Mythos-class, complete with safeguards against risky uses like cybersecurity that some users found too restrictive.

Specialists warned that Mythos systems in malicious hands could speed up complex cyberattacks, especially against banking sectors with intricate legacy networks.

Anthropic said it collaborated with the US government on safety measures before the Fable release and noted rival models also reveal minor code issues.

The order requires sudden shutdown of Fable 5 and Mythos 5 for every customer to stay compliant, while other Anthropic models remain available, the company explained.

Anthropic believes a misunderstanding occurred and is working to restore access promptly.

If applied industry-wide, this standard would likely stop all new frontier model launches, the firm warned.

Amazon's AWS unit reported late Friday that Anthropic requested revocation of model access for users everywhere.

A US official verified that the Commerce Department issued the export control order blocking foreign national access to both models.

Former White House official Dean Ball commented on X that the directive appears to bar non-Americans, even those inside the US, from the latest Anthropic systems.

This could require proof of citizenship for model access, Ball noted.

Key Anthropic staff including co-founder Chris Olah, researcher Andrej Karpathy, and philosopher Amanda Askell were born abroad. Reuters could not confirm their citizenship, and a company spokesperson declined to comment on potential access loss for such employees.