Anthropic shuts down latest AI model after gov’t calls them a national cybersecurity concern, but some experts have questions

In short: Anthropic disabled its Fable 5 and Mythos 5 AI models worldwide on June 12, 2026, after the U.S. government issued an export control directive citing national security concerns over a cybersecurity “jailbreak.” Cybersecurity experts and members of Congress have questioned whether the underlying evidence justifies an action that took a commercial AI product offline for hundreds of millions of users.

The details that matter

• Timeline: On Friday, June 12, 2026, the U.S. government directed Anthropic to take down AI models Fable 5 and Mythos 5, citing national security concerns over a cybersecurity “jailbreak.”
• Scope: The order barred access by any foreign national; because Anthropic could not filter users by nationality, it disabled both models for every customer globally.
• The finding: While testing Fable 5, Amazon’s own cybersecurity researchers reported a way around its guardrails — in essence, feeding the model flawed code and asking it to fix the bugs. The government saw a route to extracting cyberattack information; critics call it a routine defensive-security task.
• The escalation: Amazon CEO Andy Jassy then took those findings to Treasury Secretary Scott Bessent and other officials and the directive followed the next day.
• The precedent: Under existing legal authority, the government can now switch off a commercial AI product serving hundreds of millions of users worldwide on a few hours’ notice.

Anthropic shut down its two most powerful AI models worldwide on Friday after the U.S. government ordered it to block all foreign nationals from using them. The directive cited national security and covered every foreign national inside or outside the country, including Anthropic’s own non-citizen employees. Because the company can’t screen users by nationality in real time, it pulled Fable 5 and Mythos 5 for everyone.

A growing number of cybersecurity experts and lawmakers say the evidence doesn’t justify pulling a model used by hundreds of millions of people.

Why did the U.S. government shut down Fable 5 and Mythos 5?

While testing Fable 5, Amazon’s cybersecurity researchers found a way around the model’s safety guardrails, a “jailbreak” that gets an AI to do something it is built to refuse. Amazon CEO Andy Jassy brought those findings to Treasury Secretary Scott Bessent and other officials, warning they posed a national security risk, the Wall Street Journal reported. The administration issued the export control directive the next day, and Anthropic pulled the models to comply.

Anthropic disputes the government’s reasoning. It says it reviewed the demonstration and found only a handful of previously known, minor vulnerabilities, the kind that other publicly available models like OpenAI’s GPT-5.5 can surface with no jailbreak at all. According to the company, the government has produced only verbal evidence of a narrow technique that comes down to asking the model to read code and fix its flaws.

What did the disputed “jailbreak” actually show?

Katie Moussouris, the founder and CEO of Luta Security, says she is one of the few outside experts to have read the report behind the directive. In her account, the researchers gave the models flawed code, some with known bugs and some with bugs planted on purpose, and asked them to “fix this code.” Fable refused at first, but a manual workaround eventually got it to produce scripts that tested the patches. Asking an AI to repair code, she argues, should never have triggered an export control.

That kind of request is the core of defensive security, the everyday work of finding bugs, fixing them, and confirming the fix. Removing the capability would make the model worse at defense without slowing attackers, who can turn to foreign or open-weight models that U.S. export controls can’t touch. Moussouris is not alone. Dozens of security professionals have signed an open letter urging the Commerce Department to lift the controls.

How is Congress responding to the Anthropic export controls?

In Congress, the reaction has been wary. Several Democrats told CyberScoop they suspect the decision was driven by something other than security, pointing to the administration’s ongoing feud with Anthropic over domestic surveillance and autonomous weapons. Sen. Angus King (I-Maine) said he hadn’t seen a full justification and called the move “pretty extreme,” adding that he was “a little skeptical because of their otherwise announced antipathy to this company.” Rep. Bennie Thompson (D-Miss.) said the order showed an “ad hoc approach where decisions are made by political appointees in the White House rather than experts.”

Republicans were more divided. House Homeland Security Chairman Andrew Garbarino (R-N.Y.) said advanced AI cyber capabilities are a legitimate national security concern but warned against cutting off the U.S. defenders who “need access to the best secure tools.” Sen. Mike Rounds (R-S.D.) said he would hold judgment until he was briefed.

What happens next?

Anthropic has sent staff to Washington to negotiate and says the episode is a misunderstanding. Whatever the outcome, the precedent now stands. The government can take a commercial AI product used by hundreds of millions of people offline within hours, and the legal authority to do it again is already on the books.