June 12 (Reuters) - Anthropic announced on Friday that it will "immediately disable" its most advanced artificial intelligence models for all users after the U.S. government instructed the company to suspend access for foreign nationals, citing national security concerns.

According to a company statement, Anthropic received an export control order requiring it to halt access to its Fable 5 and Mythos 5 models for all foreign nationals. The directive did not include detailed explanations regarding the specific security risks involved.

Anthropic said it believes the government is concerned about a possible method of bypassing, or "jailbreaking," built-in safeguards designed to prevent Fable 5 from being used to detect software vulnerabilities.

The decision arrives at a moment when earlier tensions between officials in the Trump administration and Anthropic, which is preparing for an initial public offering, had appeared to be easing within parts of the federal government.

Relations between Anthropic and U.S. authorities deteriorated earlier this year after the company declined to permit its AI systems to be used by the military for domestic surveillance or fully autonomous weapons. In response, the government placed Anthropic on a supply chain blacklist scheduled to take effect later this year.

This move represents a significant expansion of U.S. efforts to limit the AI capabilities of foreign adversaries. Historically, export controls have targeted the semiconductor chips and technical infrastructure behind AI systems, rather than restricting direct access to the AI models themselves.

Anthropic stated that officials provided only "verbal evidence of a potential narrow, non-universal jailbreak."

"We do not believe that identifying a limited potential jailbreak justifies recalling a commercial model used by hundreds of millions of people," the company said.

The directive and Anthropic’s response underscore rising friction between AI developers and regulators over how to evaluate and manage the risks posed by so-called jailbreaks—techniques used to circumvent model safeguards.

Earlier in the week, Anthropic had publicly supported stronger U.S. oversight of AI, including mechanisms to block models that pose unacceptable risks. However, it argued that Friday’s government action did not align with principles of balanced and evidence-based regulation.

Kirsten Davies, the Pentagon’s chief information officer, stated on X that the Defense Department supports prioritizing national security considerations.

"Some priorities outweigh revenue cycles, media narratives, and IPO ambitions. America First. Always," Davies wrote.

Anthropic confidentially filed for a U.S. initial public offering last month, moving slightly ahead of competitor OpenAI in the race to enter public markets.

Earlier this week, the company introduced a new AI system called Claude Fable 5, marking the debut of what it describes as its "Mythos-class" tier of capabilities. The model includes guardrails that prohibit use in sensitive domains such as cybersecurity—restrictions that some users have criticized as overly restrictive, according to Anthropic.

Experts have warned that Mythos-class systems, if misused, could significantly accelerate advanced cyberattacks, particularly in industries like banking that depend on complex, interconnected, and often aging technological infrastructure.

Anthropic said it had collaborated with U.S. government agencies and other partners on safety evaluations prior to launching Fable 5. It added that competing AI models demonstrate similar capabilities in identifying minor coding flaws.

"As a result of this order, we must immediately disable Fable 5 and Mythos 5 for all customers to comply. Access to our other models will remain unchanged," the company stated.

Anthropic indicated that it believes the situation stems from a "misunderstanding" and said it is working to reinstate access to the models as quickly as possible.

"If this interpretation were applied across the AI sector, it would effectively stop the release of new frontier models industry-wide," the company added.

Amazon Web Services (AWS) said late Friday that Anthropic had requested the removal of access to the affected models for all users across every region.

A U.S. official confirmed that the Commerce Department had issued an export control order suspending foreign nationals’ access to Fable 5 and Mythos 5.

Dean Ball, a former White House official involved in drafting the administration’s 2025 AI Action Plan, wrote on X that the directive appears to restrict all non-U.S. citizens from using Anthropic’s newest models, including those residing within the United States.

"This suggests users may need to verify U.S. citizenship to access Anthropic’s models," Ball stated.

Several prominent Anthropic employees—including co-founder Chris Olah, AI researcher Andrej Karpathy, and philosopher Amanda Askell—were born outside the United States. Reuters could not confirm their current citizenship status, and an Anthropic spokesperson declined to clarify whether affected staff members would lose access to the company’s AI systems.