Anthropic halts new AI models after US order blocks foreign access

US export control order

Anthropic was compelled to immediately shut off access to its newest AI models late this week after receiving an export control directive from the U.S. Department of Commerce, the company disclosed in a statement. The order, citing national security prerogatives, requires the suspension of Fable 5 and Mythos 5 for any foreign citizen, whether inside or outside the United States, including foreign employees of Anthropic. The company said it had received only verbal evidence of a possible vulnerability and no detailed specifics about the security concern.

The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all of our customers to ensure compliance.

Fable 5, a variant of the larger Mythos system, had been launched only days earlier and competes with OpenAI’s ChatGPT and Google’s Gemini. Anthropic stressed that access to all its other models remains unaffected.

Anthropic's response and jailbreak claims

The company understands that the government believes it identified a method to bypass, or jailbreak, safeguards that would prevent Fable 5 from being misused for cybersecurity activities. Anthropic acknowledged that its internal review found a few minor, previously known vulnerabilities, but argued that similar weaknesses exist in other public models without requiring advanced circumvention techniques. The firm also highlighted that it had collaborated with the U.S. government, the UK’s AI Safety Institute, independent organisations and internal teams for thousands of hours of testing before launch.

We do not agree that the discovery of a potential limited jailbreak should constitute a reason for withdrawing a commercial model deployed for hundreds of millions of people.

Anthropic said it was working to clarify the situation and restore access as quickly as possible. It noted that just two days before the directive, it had publicly called for stricter U.S. oversight of AI, including the ability to block models deemed too dangerous, but it believed Friday’s measure did not follow principles of fair, evidence-based regulation.

Strained relations with the Trump administration

The order deepens an already tense relationship between the San Francisco-based company and the Trump administration. Earlier this year, Anthropic refused to allow the U.S. military to use its models for domestic surveillance and fully autonomous weapons systems. In March, the Pentagon labelled the company an unacceptable supply-chain risk, a blacklisting that is expected to take effect later this year. The current directive marks a further escalation, moving U.S. export controls beyond chips and hardware tools to directly restricting foreign access to advanced AI software.

Expert warnings and industry implications

The suspension could hamper safe development and testing of powerful models, warns Gina Neff, a professor of responsible AI at Queen Mary University of London.

We are in uncharted territory at this moment. People in the AI industry have warned us that these tools are improving very fast and we need to be able to develop capabilities to protect ourselves.

Kirsten Davies, the Pentagon’s chief information officer, backed the priority of national security.

National security must take precedence and some matters are more important than company revenues, media attention or pre-IPO valuations.

The clash between a leading AI developer and regulators highlights the growing tension over how to assess and govern risks from increasingly capable AI models, particularly as Anthropic itself pursues a stock market listing.