U.S. Government Orders Anthropic to Pull Fable 5 and Mythos 5 Offline Over Jailbreak Concerns. Anthropic Says the Risk Is Overstated.

Since Anthropic's escalating conflict with the Trump administration began earlier this year, when the Department of Defense labeled the company a 'supply chain risk' and barred contractors from using Claude, the tension has now produced its sharpest flashpoint yet.

At 5:21 p.m. ET on Friday, June 12, the U.S. government delivered an export control directive ordering Anthropic to suspend all access to Fable 5 and Mythos 5 by any foreign national, including foreign national Anthropic employees, whether inside or outside the United States. Anthropic published the full directive context in a statement on its website the same evening.

Rather than attempt to verify the nationality of every active user, Anthropic cut access to both models entirely. All other Claude models remain available as of June 12, 2026.

What Fable 5 Actually Is

Fable 5 launched publicly earlier this week. It is a restricted version of Mythos 5, the same underlying model that debuted in April under Project Glasswing, a partnership with Apple, Google, and Microsoft to find and fix vulnerabilities in critical internet infrastructure, according to ZDNET.

The key word is restricted. Anthropic built Fable 5 specifically with guardrails blocking responses in high-risk cybersecurity, biology, and chemistry territory. When users hit those limits, the model downgrades to Opus-level intelligence. For most flagged categories, users are told the downgrade is happening. For a narrower category—research into advanced chip design and frontier AI models—ZDNET reported the downgrade happened silently, a disclosure buried 319 pages deep in the Fable and Mythos System Card. That hidden throttle sparked its own backlash earlier Friday around transparency and research trust.

What the Government Claims, and What Anthropic Says About It

The government's letter, per Anthropic's own statement, did not provide specific details about the national security concern. Anthropic says its understanding is that the government believes it found a method to jailbreak Fable 5's safeguards.

Anthropic's published response is direct: it reviewed a demonstration of the specific technique and found it produced "a small number of previously known, minor vulnerabilities." The company says those vulnerabilities are replicable by other publicly available models, including GPT-5.5, without requiring a bypass of any kind.

Anthropic's statement also noted that in the weeks before launch, it red-teamed Fable 5 for thousands of hours with the U.S. government, the UK AI Safety Institute, multiple third-party organizations, and internal teams. No tester found a universal jailbreak, meaning one that could broadly unlock a wide range of dangerous capabilities.

The company's argument: the jailbreak shown is non-universal, narrow in scope, and not unique to Fable 5. Holding Fable 5 to a standard of zero jailbreak vulnerability, Anthropic wrote, would "effectively halt all new frontier model deployments" across the entire industry.

The Case for Taking the Government's Concern Seriously

The strongest counter-argument is not paranoia. Fable 5 was explicitly built on Mythos, which was restricted to select organizations for exactly the reason ZDNET described: a tool powerful enough to find unknown vulnerabilities to fix them is also powerful enough to find them to exploit them. Making that capability broadly available to hundreds of millions of users, even with guardrails, is a categorically different risk profile than keeping it inside Project Glasswing.

Sally Vincent, senior threat research engineer at security analytics firm Exabeam, told ZDNET that jailbreak-resistance claims "should be viewed with appropriate caution" because they "represent a point-in-time assessment" and "attackers continuously adapt." Anthropic's own statement acknowledges that perfect jailbreak resistance is not currently possible for any provider, and that universal jailbreaks will "eventually be found."

Where the government's position gets harder to defend is in the specifics. Anthropic says it reviewed the actual demonstration and found it narrow. If the government has evidence of a more serious, broader bypass, it has not shared that with Anthropic or the public. The letter provided no specific details.

The Larger Fight

This is not happening in a vacuum. The DoD's 'supply chain risk' designation earlier this year, a label historically applied to foreign adversaries, is the backdrop. Anthropic's lawsuit against the Trump administration over that blacklisting is still active, according to Wired. The company is simultaneously complying with Friday's order and characterizing it as a misunderstanding.

Wired noted the government's order came just days after Anthropic publicly released Fable 5, which Anthropic had said was conducted in collaboration with the U.S. government. Receiving a shutdown order for a model the government helped vet is a notable reversal, and neither the White House nor the relevant agency had issued a public statement as of the time Anthropic published its response Friday evening.

The open question now is whether the government will share the specific jailbreak evidence with Anthropic under any disclosure process, and whether Anthropic can patch the identified technique quickly enough to make a case for restoring access while its DoD lawsuit remains unresolved.