Anthropic Pushes Back Publicly as Fable 5 and Mythos 5 Stay Offline, Calls Shutdown Order a Misunderstanding

Since the U.S. government's June 12 directive forced Anthropic to pull Fable 5 and Mythos 5 offline for all users worldwide, the company has sharpened its public criticism of how the order was handled, even as it remains in compliance.

Anthropic is not contesting the government's authority to restrict AI exports. Its objection is procedural and technical.

According to Economic Times, Anthropic said it "did not receive detailed evidence outlining the specific national security concerns" behind the directive. The company told the government it reviewed a demonstration of the alleged jailbreak method and concluded it exposed only "a limited number of previously known and minor issues" — vulnerabilities that are, in Anthropic's own words, "widely available in other publicly accessible AI systems and are routinely used by cybersecurity professionals."

The specific technique at issue involves prompting Fable 5 to analyze codebases and flag software vulnerabilities. Anthropic's position is that this is a standard cybersecurity workflow, not a novel weapons-grade exploit.

"We believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts," the company said in a statement, reported by both Livemint and Economic Times. "This action does not adhere to those principles."

Anthropic's strongest argument rests on the paper trail behind Fable 5's release. The company says it ran pre-release safety evaluations in collaboration with U.S. and UK government bodies, independent organizations, and internal teams, involving thousands of hours of red-teaming, according to Economic Times. That work produced no finding of a "universal jailbreak" — defined by Anthropic as a technique capable of broadly bypassing the model's protections. If accurate, that's a meaningful data point. The government bodies now raising alarms were, at least in part, involved in the pre-release review. The directive does not explain why the same concern wasn't flagged then.

The government's strongest case for the shutdown is straightforward: a model powerful enough to autonomously analyze codebases for vulnerabilities, if jailbroken even partially, could accelerate cyberattack development in adversarial hands. The government's interest in restricting foreign-national access to frontier AI models is not paranoia. It is exactly the kind of national security judgment the executive branch is supposed to make. Anthropic's own characterization of Mythos 5 as a model it has "tightly limited access to due to cybersecurity fears" (per Livemint) suggests the company itself believes these tools carry serious risk.

The problem is the Commerce Department has said nothing publicly. It did not respond to requests for comment from the Associated Press as of the time these reports were published on June 13. No detailed technical finding has been released. That silence makes it impossible to independently evaluate whether the jailbreak claim is valid or whether Anthropic's rebuttal holds up.

The directive lands ten days after President Trump signed an executive order establishing a federal framework to vet national security risks from advanced AI systems for up to 30 days before public release, according to Livemint. Participation by AI developers is voluntary under that order.

The Fable 5 shutdown is a different instrument entirely — a direct export control directive, not the voluntary vetting process. That distinction matters. If developers can't trust that voluntary pre-release cooperation will protect them from abrupt mandatory shutdowns, the incentive to participate in the voluntary framework weakens considerably.

Fable 5, which Anthropic released broadly this week, and Mythos 5, the more restricted advanced version, are both disabled for all users globally as of June 13, 2026. Economic Times confirmed that access to all other Anthropic models remains unaffected. The company's enterprise partnerships, including the TCS deal announced June 11 deploying Claude to more than 50,000 employees, run on different models and have not been reported as impacted.

Anthropic says it is "working to restore access as soon as possible" and has characterized the situation as a misunderstanding it hopes to resolve quickly.

The unresolved question is whether the Commerce Department will publish the technical basis for its jailbreak claim. Without that, there is no independent way to determine whether Anthropic's rebuttal is accurate, whether the government identified a real gap in the pre-release testing process, or whether the shutdown sets a replicable precedent for halting advanced AI deployments on undisclosed evidence.