White House restricts Anthropic models over China access fears

The White House has put Anthropic’s most advanced models under export controls after suspicions that a China-linked group may have accessed Mythos, turning a model-safety dispute into a national security fight. Anthropic said it received the directive at 5:21 p.m. ET on June 12 and disabled Fable 5 and Mythos 5 for all customers, while saying its other models were unaffected.

The order required Anthropic to suspend access to the two models for any foreign national, inside or outside the United States, including foreign-national Anthropic employees. That broad restriction landed just days after Anthropic had announced Fable 5 and Mythos 5 as state-of-the-art systems, and it immediately forced the company to pull them back from general use.

At the center of the dispute is Mythos, which Anthropic had tightly controlled after launching it in April. The company limited access to select companies using it to help find security flaws, a move Anthropic said reflected concerns that the model’s cyber capabilities could be misused by hackers. Semafor reported that the White House action was partly driven by suspicions that a China-linked group had accessed Mythos, but it remained unclear which organization had gained access or how it did so. If Chinese officials or affiliates had obtained the model, analysts warned, they could try to reverse-engineer it through distillation.

Anthropic said the government did not give specific details about its national-security concern. The company’s understanding was that officials believed they had learned of a jailbreak method for Fable 5, not Chinese access, and Anthropic said a review of the technique showed only a small number of previously known, minor vulnerabilities. The company also said it had spent thousands of hours working with the U.S. government, the U.K. AI Security Institute, private third parties and internal teams to red-team Fable’s safeguards. Anthropic said Fable 5 already had guardrails that prevented it from answering questions about cybersecurity and biology.

The political fallout widened as White House adviser David Sacks said the government had received a warning that Fable 5 could be jailbroken and that Dario Amodei had allegedly refused to fix the issue. A person close to the White House said Amazon had alerted officials about the jailbreak and that Andy Jassy had been in contact with administration officials. Anthropic said the White House did not raise Chinese access in its conversations about the jailbreak and export controls.

The episode now gives Washington a fresh case study in how quickly a frontier AI product can become a trade and security issue. For policymakers weighing broader restrictions, the Anthropic move suggests the debate is no longer just about model performance, but about who can access the systems, how they can be tested, and how much of the technology the government is prepared to lock down.