China access concerns could have triggered US restrictions on Anthropic's Mythos: Report

The White House imposed export controls on Anthropic's powerful Mythos artificial intelligence (AI) model partly over concerns that a China-linked group may have gained access to it, Semafor reported, citing a person familiar with the matter, in a sign of how frontier AI models are increasingly being treated as strategic assets akin to semiconductors.

The Trump administration on Friday directed Anthropic to limit access to Mythos and its consumer version, Fable 5, to US citizens. The requirement would have barred even foreign nationals working at Anthropic from using the models, prompting the company to disable access globally.

Also Read: US ban on Anthropic's Fable 5 & Mythos 5 to put Indian IT services firms at competitive disadvantage


The episode highlights a growing reality in the AI race where it is no longer just about building the most powerful models, but also about controlling who gets access to them.

According to Semafor, it remains unclear how the White House became aware of the alleged access, which China-linked entity may have obtained the model, or how access was gained. But if Chinese entities had access to Mythos, it could raise concerns around espionage and "distillation", which is the process of reverse-engineering advanced AI systems to replicate their capabilities.

Anthropic launched Mythos in April with tightly controlled access, limiting its use to a select group of companies for cybersecurity applications such as identifying vulnerabilities in software systems. The company has previously warned that the model's ability to find bugs in code could be exploited by malicious actors.

The San Francisco-based startup, which has confidentially filed for a US initial public offering, had held back Mythos from wider release because of concerns over its hacking capabilities. Earlier this week, it rolled out Fable, a public version of the model equipped with cybersecurity safeguards.

Fresh details emerged over the weekend when White House AI adviser David Sacks said on X that the administration had been alerted to a jailbreak of Fable 5 — a technique that bypasses an AI model's safety guardrails.

Sacks alleged that when the administration informed Anthropic about the issue, chief executive Dario Amodei downplayed the risks and declined to address them. "In this case, Anthropic prioritized the continued offering of the consumer model over safety," Sacks wrote.

A person close to the White House told Semafor that Amazon had informed the government about the jailbreak and that chief executive Andy Jassy had been in touch with administration officials. The Information first reported Jassy's outreach.

Also Read: Explained | Why Anthropic’s Mythos is spooking bankers and what it means for global finance

Anthropic, however, tols Semafor concerns regarding Chinese access to Mythos were not raised during discussions with the White House around the jailbreak and export controls. The company also noted that it blocks access to its products from within China.

Meanwhile, senior Anthropic technical staff are in Washington to meet White House officials next week in an effort to resolve the dispute, Axios reported on Sunday, citing a source close to the company. Reuters also reported that Anthropic staff have also held virtual meetings with administration officials since Friday.