Damage control? Anthropic rushes to Washington amid White House ban on top AI models ahead billion-dollar IPO

Anthropic has dispatched senior technical staff to Washington to hold face-to-face talks with White House officials, as the company races to resolve a deepening dispute that has taken its most advanced artificial intelligence models offline for users across the globe, a person familiar with the development told Axios.

How the export control crisis unfolded

The crisis began on 12 June, when the US Department of Commerce issued an export control directive ordering Anthropic to block access to its two newest models, Fable 5 and Mythos 5, for foreign nationals both inside the United States and abroad. Because reliably filtering users by nationality was technically unfeasible, Anthropic took the decision to disable both models entirely for all users worldwide

The shutdown came just three days after the models had been made publicly available, on 9 June, in what Fortune described as a "considerable step" for the lab, arriving little more than a week after Anthropic had confidentially filed for IPO paperwork.

What are Mythos and Fable, and why do they matter?

Mythos is Anthropic's most powerful frontier model. When the company first announced it in April, it acknowledged the model was too capable in the domain of cybersecurity to release safely, citing its significantly enhanced ability to identify software vulnerabilities.

Rather than a public rollout, Mythos was made available to a small number of trusted organisations, mostly US technology corporations, to help identify and patch weaknesses in critical digital infrastructure.

Fable 5 is built on the same underlying model but incorporates additional safeguards designed to prevent it from being used for cybersecurity purposes. It was this version that Anthropic released to the public on 9 June, before shutting it down almost immediately under US government pressure.

The hidden restriction that ignited a backlash

Within hours of Fable 5's public release, significant criticism began circulating among AI researchers, developers and policy experts. The backlash centred on a disclosure buried within the model's 319-page system card, a document providing detailed safety information, which revealed that Fable 5 would quietly downgrade its own responses when it detected queries related to advanced AI development work, such as building the infrastructure used to train large AI models.

In practice, this meant a user could ask the model a question, receive a deliberately weakened answer, and have no indication that the model was withholding anything.

Unlike Fable's other restrictions, such as those covering cybersecurity and biology, which visibly redirect users to a less capable model with a clear notification, the system card stated that this particular limitation is "not visible to the user." The model would still respond, but would deploy "interventions to limit Claude's effectiveness" without any disclosure to the person asking.

Anthropic estimated the restrictions would affect roughly 0.03 per cent of traffic, and defended the approach by stating that "enforcing this restriction through our safeguards avoids accelerating the actors most willing to violate these terms."

Anthropic and Trump administration: an escalating confrontation

The export controls did not emerge from a vacuum. Since early 2025, relations between Anthropic and the Trump administration have been deteriorating steadily. Administration officials have accused the company of producing "woke AI" and labelled chief executive Dario Amodei an “ideological lunatic.”

Early tensions centred on disagreements over AI regulation and semiconductor export policy. The conflict intensified when Anthropic declined to permit the Pentagon to use its models for domestic surveillance operations and fully autonomous weapons systems. The Department of Defense responded by threatening to designate Anthropic a "supply chain risk," a classification that would have compelled military contractors to cut ties with the company entirely.

Talks under way, but trust remains fragile

Despite the gravity of the situation, administration officials have claimed that Anthropic has not engaged in a serious manner. However, a source close to the company says that Anthropic's technical staff have held virtual meetings with White House officials since the administration made its initial approach on Friday, with the in-person Washington visit representing a significant escalation in the company's efforts to find a resolution.

The stakes are high. With an IPO filing already under way and its flagship models suspended globally, Anthropic faces mounting pressure to demonstrate that it can navigate the intersection of AI safety, government regulation and commercial viability, all at once.