The Trump administration imposed export restrictions on Anthropic's artificial intelligence models — Fable 5 and Mythos 5 — after receiving data on possible access to Mythos by a group linked to China. This incident became a key trigger for the White House, although officials have not publicly confirmed this version.

At the same time, reports emerged about bypassing Fable 5's safety mechanisms, which also served as a basis for the restrictions. Anthropic itself insists that it was a "narrow" and non-universal method that did not give the model unique capabilities. The company denies that the issue of "Chinese access" to Mythos was raised in negotiations with authorities and emphasizes that it prohibits access to its products from China.

Events Preceding the Restrictions

On June 11, Amazon CEO Andy Jassy informed the presidential administration about the possibility of bypassing Fable 5's protections in response to a request to review the model. U.S. Secretary of Commerce Howard Lutnick sent a letter to Anthropic CEO Dario Amodei, stating that Fable 5 and Mythos 5 were subject to export controls outside the U.S. and for all foreign individuals within the country. According to sources, officials tried to persuade Anthropic to voluntarily suspend access, but the company refused. Sources close to the startup claim the White House gave only 90 minutes to shut down the models without explaining the nature of the threat.

On June 13, White House advisor David Sacks stated that the administration received a warning about a possible bypass of Fable 5's protections from a partner trusted by both the authorities and Anthropic. "The administration hopes that Anthropic will fix the security issue, the export controls will be lifted, and Fable will return to public access. The ball is in Anthropic's court," Sacks wrote.

Anthropic's Position

Anthropic claims that the authorities' letter did not contain specific details about risks to national security. The company reviewed the demonstration of the technique to bypass Fable 5's protections and concluded that it allows finding "a small number of previously known minor vulnerabilities." Anthropic adds that other publicly available models can identify the same issues without bypassing protections.

Before launching Fable 5, the company spent "thousands of hours" testing protections jointly with the U.S. government, the UK's AI Safety Institute, and third-party organizations. Testers have not yet found a universal way to mass-bypass the model's restrictions. "We consider this a misunderstanding and are working to restore access as soon as possible," Anthropic stated.

Context and Consequences

On June 9, Anthropic introduced Fable 5 and Mythos 5. Fable 5 is a Mythos-class model for public access, while Mythos 5 uses the same base model but with relaxed restrictions in certain areas, intended for cybersecurity defenders and critical infrastructure providers under Project Glasswing. On June 2, the company announced the expansion of Project Glasswing to approximately 150 organizations from over 15 countries, which found more than 10,000 high and critical severity vulnerabilities.

Following the imposition of export restrictions, tokens of decentralized AI projects Venice and Morpheus rose by 14% and 21% respectively, demonstrating the market's reaction to tighter control over centralized AI models.

My comment: This incident highlights the growing tension between national security and openness in AI development. Restrictions imposed without clear evidence could slow innovation and increase fragmentation of the global AI market. The rise in tokens of decentralized projects signals that investors see them as an alternative less susceptible to political pressure.