White House Fears China-Linked Group Accessed Anthropic's Mythos AI Model

The White House's decision to impose export restrictions on Anthropic's Mythos AI model was driven in part by suspicions that a China-linked group had gained access to the system, according to a report from Semafor published Saturday.

The report adds a significant national security dimension to what the administration had previously framed as a safety and security concern over the model's capabilities.

If confirmed, unauthorized Chinese government access to Mythos 5 or its companion model Fable 5 would represent a serious national security risk, analysts say. One concern is the potential for reverse engineering through a technique known as distillation, in which a less capable "student" AI is trained on outputs from a more advanced model in order to replicate its behavior.

The White House has not confirmed the Semafor report. Trump advisor David Sacks addressed the export restrictions in a post on X but made no mention of China, focusing instead on reported vulnerabilities that allow Fable and Mythos to be jailbroken — a characterization Anthropic has disputed.

Anthropic did not reply to a request for comment. A spokesperson told Semafor that the Chinese government was not raised during the administration's discussions with the company around export controls.

Sen. Mark Kelly (D-Ariz.) on Sunday offered rare bipartisan support for the administration's position, backing the decision to urge Anthropic to suspend access to its latest models. "Mythos and some of their other models from AI companies are incredibly capable, and some of the capabilities that these models have," Kelly said, underscoring broader congressional concern about frontier AI systems regardless of party affiliation.

The reported breach, if accurate, would not be the first time Anthropic's most powerful model was accessed outside of authorized channels. The company has maintained that Mythos is too dangerous for public release, yet a Discord group reportedly had access to the model for approximately two weeks before Anthropic discovered the exposure and terminated it.

That earlier incident highlights the difficulty of controlling access to frontier AI systems even when developers intend to keep them restricted — a challenge that has accelerated calls in Washington for formal export control frameworks covering advanced AI models.

The episode also lands against a fraught legislative backdrop. Congress has been unable to extend Section 702 of the Foreign Intelligence Surveillance Act, a key legal authority used for foreign intelligence collection. Sen. Shelley Moore Capito (R-W.V.) on Sunday publicly criticized fellow lawmakers over the impasse, urging them to "stop playing the politics" as both chambers remain deadlocked on renewal legislation.

The convergence of the FISA stalemate and the Mythos access concerns illustrates the degree to which Washington's national security apparatus is grappling simultaneously with gaps in traditional surveillance law and emerging threats tied to advanced AI systems.

Whether the China-access report accelerates legislative action on AI export controls — or remains disputed by the parties involved — is likely to define the next phase of the Mythos controversy.

Disclaimer