U.S. blocks foreign access to Anthropic’s Claude Fable 5, Mythos 5

On Friday the U.S. government issued an emergency export control directing Anthropic to immediately block access to its Claude Fable 5 and Claude Mythos 5 models for all foreign nationals, citing a potential jailbreak that officials view as a national security risk. Anthropic complied with the order but disputed the severity of the reported vulnerability.

The directive prohibits any foreign national from using the two models whether they are inside or outside the United States. Because the restriction applies to all foreign users, Anthropic temporarily disabled the models for its entire customer base to ensure compliance. Fable 5 had been released publicly days earlier while Mythos 5 was available only to select partners.

U.S. officials informed Anthropic they had identified a method to bypass safety guardrails on the publicly released Fable 5. According to Anthropic, government representatives provided verbal descriptions of a narrow, non-universal jailbreak that involved asking the model to read a specific codebase and fix software flaws. The administration characterized the issue as a national security matter and used emergency export controls to limit access.

Anthropic reviewed a demonstration of the technique and concluded the vulnerability appeared relatively simple. The company asserted the capability shown could be replicated with other publicly available models, including GPT-5.5, and that the government had not provided full technical evidence in writing. Anthropic wrote that applying the government’s standard across the industry would, in its view, halt new deployments of high-capability models.

Anthropic said access to its other models is not affected and the company is working to restore services as quickly as possible. The firm also indicated it is prepared to work with regulators and testers to address specific technical issues.

Mythos 5 was designed with fewer guardrails and performs strongly in tasks related to cybersecurity and code analysis. That class of model had been distributed under restricted access to researchers and partners. Security researchers and government testers evaluating Mythos-class systems under limited conditions reported the potential bypass that prompted the export control action.

David Sacks, co-chair of the President’s Council of Advisers on Science and Technology, posted on X that a trusted partner testing Fable disclosed a jailbreak and that the administration asked Anthropic’s CEO, Dario Amodei, to fix the issue or remove the model from availability. The post said Anthropic declined the government’s request and that officials acted reluctantly in issuing the export control.

The measure follows earlier tensions between Anthropic and the U.S. government. The companies previously disagreed over an expanded data-access agreement, and the Department of Defense labeled Anthropic a supply-chain risk, a designation the company has contested in court. Negotiations since then included discussions about government access to Anthropic’s Mythos models.

Public reaction to Fable 5’s brief public release was mixed. Users noted strong coding performance but also reported high token consumption and inconsistent behavior in some research tasks. Government officials and Anthropic have indicated they will pursue further technical reviews and fixes before any resumption of public or partner access for the two models.