Anthropic’s latest fight with the Trump administration has grown into a broader dispute touching national security, export controls, and the company’s standing in the AI market.

The controversy centers on Anthropic’s most advanced models, Mythos and Fable, which were taken offline after the government imposed tighter export restrictions. Administration officials say the company mishandled security concerns tied to the models, while Anthropic argues it had already taken steps to address the issues and had government approval to deploy at least one of the systems.

The conflict intensified after Amazon CEO Andy Jassy contacted Treasury Secretary Scott Bessent to warn that Anthropic’s models could be jailbroken. Administration officials say the company was aware of that risk and proceeded anyway. Anthropic disputes that characterization, saying it had explicit government approval to launch Fable.

The White House has also said it previously threatened Anthropic with export controls after learning that a version of its Mythos model had been made available to an entity in a foreign country with ties to the Chinese Communist Party. A source close to Anthropic said the company has worked closely with the government on broader access to Mythos and had revoked access in the telecom-related case before any threat of export controls surfaced.

Still, the administration’s frustration appears to have spread beyond the technical concerns. One official described the company as failing to communicate effectively with the government, saying Anthropic did not fully appreciate the administration’s perspective. Another source familiar with the White House’s thinking said the two sides simply speak different languages and have not figured out how to bridge the gap.

That communication breakdown comes on top of earlier friction between Anthropic and the Pentagon, which one White House official said was separate from the current dispute but reflected a similar dynamic. The official said the administration had wanted to avoid the outcome but felt compelled to act.

The result was a blunt policy response. On Friday night, the government moved ahead with stringent export controls that effectively forced Anthropic to shut down access to the models. Administration officials portrayed the move as necessary to address a serious national security concern, while Anthropic said it had been caught in a situation where its hands were effectively tied.

The episode also carried reputational consequences. After the models went offline, Anthropic published a blog post pushing back on the Amazon report and brought in a cybersecurity expert whom administration officials viewed as politically hostile. That move, according to sources cited in the report, worsened the optics at a sensitive moment.

Anthropic has become one of the most vocal frontier AI labs warning about the dangers of advanced systems, especially in cyber applications. That posture has made it a prominent voice in the policy debate, but also one that has frequently unsettled the Trump administration and parts of the public.

The White House had recently worked to improve ties with the company after the earlier Pentagon dispute. But the latest episode has put that thaw at risk. An administration official said the immediate crisis had been contained, but warned that longer-term problems remain.

Several follow-up meetings are now planned. Commerce Department officials are scheduled to meet Monday with Anthropic staffers Logan Graham, Dave Orr, and Nicholas Carlini. Separate discussions are also set with the CIA and White House science adviser Michael Kratsios as the administration tries to ensure compliance with a new cyber executive order.

For now, the dispute underscores how quickly AI policy can turn into a national security fight, especially when technical risks, export rules, and political mistrust collide.