White House AI curbs reportedly tied to China access concern
Semafor says fears that a China-linked group reached Anthropic’s Mythos helped drive new White House export restrictions.
By Hana Yoshida · Markets Reporter
3 min read
The White House’s move to restrict exports of Anthropic’s Mythos AI system was partly driven by concern that a China-linked group may have accessed it, Semafor reported. The concern matters because U.S. officials have treated Anthropic’s most advanced models as sensitive technology with potential national security implications.
Semafor reported that the White House suspected a group connected to China had access to Mythos. The White House has not publicly confirmed that account, according to The Verge.
The risk, as described by The Verge, is that access by the Chinese government to Mythos 5 or Fable 5 could expose powerful AI capabilities to a strategic rival. The Verge also reported that such access could raise fears of model copying through a technique known as distillation.
IBM describes knowledge distillation as a process in which a smaller or less capable “student” model is trained using outputs from a stronger model, allowing it to imitate some of the stronger system’s behavior. In an AI policy fight, that makes access to a restricted model more than a normal security lapse: it could help another party reproduce capabilities without obtaining the original system.
White House has not cited China publicly
David Sacks, an adviser to President Donald Trump, addressed the Anthropic restrictions in a post on X, The Verge reported. Sacks did not mention China, and instead pointed to reports that Fable and Mythos could be jailbroken.
Anthropic has denied that claim, according to The Verge. The company had not responded to The Verge’s request for comment, while an Anthropic spokesperson told Semafor that China did not come up in the company’s talks with the government about export controls.
The competing explanations leave the public record incomplete. Semafor reported that China-related concerns helped shape the White House action, while the administration’s public messaging, through Sacks, focused on alleged jailbreak risks.
Anthropic’s model security is under scrutiny
The new report adds pressure on Anthropic over access controls around its strongest AI systems. The Verge reported that Anthropic has described Mythos as too powerful and dangerous for general public release.
Mythos has faced earlier access issues, according to The Verge. The company previously said a Discord group had access to the model for two weeks before Anthropic found the breach and shut it down.
The reported China concern also follows other scrutiny of Fable 5 and Mythos 5. The Verge previously reported that Anthropic cut off access to those models after a government order tied to national security concerns.
For now, the central claim remains unconfirmed by the White House. Semafor says officials feared a China-linked group had reached Mythos, Anthropic says China was not raised in its export-control discussions, and the administration’s public comments have focused on jailbreak allegations rather than foreign access.
This story draws on original reporting from The Verge.