US Directs Anthropic to Block Access to Latest Frontier AI Models

The U.S. government issued an emergency export control order Friday requiring Anthropic to immediately cut off access to its newly released Claude Fable 5 and Claude Mythos 5 models for all foreign nationals, including the firm's own employees, over alleged national security risks tied to a possible jailbreak vulnerability.

The government directive targets foreign access to new models. The Commerce Department action prohibits any foreign national from using either system inside or outside the United States. Because the order is so broad, Anthropic had to take both models offline for its entire customer base to comply. Mythos 5 reportedly features fewer guardrails and excels at finding cybersecurity exploits; it had been limited to select partners.

The directive gave no detailed explanation of the threat but indicated officials had learned of a way to bypass safeguards on the publicly available Fable 5.

Anthropic disputes the vulnerability's severity. After examining a demonstration of the method, the company determined the issues are relatively simple and already replicable on other publicly available systems, including OpenAI's GPT-5.5, without any special bypass. Anthropic characterized the government's evidence as verbal only, describing the exploit as a narrow, non-universal technique that essentially involves instructing the model to review a specific codebase and correct software flaws. Bloomberg reported the firm complied with the order while contesting its scope.

The company warns of industry-wide precedent. In its response, Anthropic argued the move sets a dangerous precedent. "If this standard was applied across the industry," the company wrote, "we believe it would essentially halt all new model deployments for all frontier model providers."

All other Anthropic models remain available. The firm said it is trying to restore access to the two systems as quickly as possible. WIRED described the episode as the latest point of tension with the Trump administration; The New York Times reported that the length of the restrictions is still unknown.

White House adviser criticizes Anthropic's response. On Saturday, David Sacks, co-chair of the President’s Council of Advisers on Science and Technology, posted on X that a trusted partner testing Fable had uncovered a jailbreak of its guardrails. Sacks wrote that the administration then asked CEO Dario Amodei to correct the flaw or remove the model from deployment, but Amodei refused. He added that Anthropic “prioritized the continued offering of the consumer model over safety” and that the export control was issued only reluctantly. Sacks said the administration hopes the company will address the issue so Fable 5 can be released again. The critique contrasts with Amodei’s own recent blog post endorsing strong AI safety measures.