Anthropic's Claude Fable 5 and Claude Mythos 5 were taken offline on June 12 after the U.S. Commerce Department invoked export control rules to address a cybersecurity jailbreak. The shutdown lasted roughly 90 minutes from the time the directive arrived, affecting every user worldwide and marking the first instance of the U.S. government using export controls to disable a commercial AI model in real time.

What Triggered the Government's Action?

The Commerce Department's concern centered on a jailbreak technique that Amazon researchers had identified. The technique worked by asking Claude Fable 5 to read a specific codebase and identify software flaws as a defensive code-review task, which bypassed the model's cybersecurity classifier that normally refuses offensive requests. The framing mattered: because legitimate security engineers use the same language when conducting defensive code reviews, any fix risked degrading the model's legitimate capability alongside blocking malicious use.

Secretary of Commerce Howard Lutnick signed the export control directive on June 12 at 5:21 p.m. ET. Within 90 minutes, both Fable 5 and Mythos 5 were offline. The legal mechanism invoked was the "deemed export" doctrine under 15 CFR 734.13, which treats the release of controlled technology to a foreign national inside the United States as an export to that person's home country. Originally written for semiconductor blueprints and technical data, the doctrine had never been applied to real-time AI inference over a cloud endpoint before.

"The Admin issued this reluctantly. It's been very surprised that Anthropic hasn't wanted to cooperate with a reasonable safety request," stated David Sacks, co-chair of the President's Council of Advisers on Science and Technology.

David Sacks, Co-Chair of the President's Council of Advisers on Science and Technology

According to Sacks, the administration had offered Anthropic a choice before issuing the directive: fix the jailbreak or voluntarily remove the model from service. CEO Dario Amodei declined both options.

Why Did the Ban Become a Global Outage?

The shutdown affected every user worldwide, not just those in the United States or from specific countries. The reason reveals a critical gap between how export control law was written and how modern AI systems operate. Anthropic could not verify user citizenship in real time at consumer scale. A billing address or email domain does not establish legal nationality under export control rules.

Rather than risk non-compliance with the directive, Anthropic disabled both Fable 5 and Mythos 5 for every customer everywhere. Claude Opus 4.8, Claude Sonnet 4.6, and Claude Haiku 4.5 remained available, as the export control targeted only the two Mythos-class models.

What Do Experts Say About the Severity of the Threat?

Anthropic disputed the government's characterization of the jailbreak risk. The company described the technique as narrow and non-universal, and said the vulnerabilities surfaced were minor, previously known, and replicable with other publicly available models, including those with no equivalent safety classifier at all.

Katie Moussouris, CEO of Luta Security and a member of the Commerce Department's Information Systems Technical Advisory Committee, reviewed the original research paper and offered a different assessment. She called the export control "heavy-handed and hasty" and argued that the bypass technique could not be patched without degrading the model's legitimate code-review capability.

"The bypass technique could not be patched without degrading the model's legitimate code-review capability, and called for restoring both models and Project Glasswing access," noted Katie Moussouris.

Katie Moussouris, CEO of Luta Security

How the Shutdown Created Diplomatic Tension

The timing of the shutdown created an acute diplomatic irony. Five days after the ban took effect, Amodei was seated at a G7 AI working lunch alongside heads of government whose citizens had just been cut off from the most capable commercial AI model available anywhere. French President Emmanuel Macron warned publicly that no country would purchase American AI if it could be switched off at any moment. The United Kingdom, Canada, Australia, and New Zealand all voiced frustration that the directive had swept up Five Eyes partners alongside adversarial nations.

A formal UK request for an exemption was turned away by the White House on June 17, the same day as the lunch. At that meeting, Amodei and Google DeepMind CEO Demis Hassabis jointly proposed a U.S.-led AI alliance of democratic nations that would coordinate advanced AI trade and safety standards while excluding China.

What Changed After the G7 Meeting?

President Donald Trump's public stance shifted noticeably after the G7 lunch. In an interview with Axios published on Friday, Trump said he no longer views Anthropic or Amodei as a national security threat. "Well, not now, but a week ago, maybe," Trump said when asked directly about the company's status. He described Amodei as "nice" and "smart" and praised the company's response to the directive.

Trump

However, a changed presidential view does not equal a lifted directive. As of Saturday afternoon, Claude Fable 5 and Mythos 5 remained offline for every user worldwide. The Commerce Department's export control directive has not been withdrawn. The Defense Department's supply-chain risk designation issued in March remains active, and the ban on federal agency use of Anthropic technology, ordered by Trump earlier this year, is still in force.

Steps to Understand the Broader Implications

• Legal Precedent: This marks the first time the U.S. government has used export control rules to disable a commercial AI model in real time, creating a new legal precedent for how deemed export doctrine applies to cloud-based AI services.
• Citizenship Verification Challenge: The shutdown exposed a fundamental technical problem: AI companies cannot verify user citizenship at consumer scale, forcing them to choose between global outages or legal non-compliance.
• Safety vs. Capability Trade-off: The jailbreak technique worked by exploiting the model's legitimate code-review capability, meaning any fix would degrade both offensive and defensive uses simultaneously.
• Geopolitical Leverage: The incident demonstrates how export controls can be weaponized to influence AI company behavior and international relations, as evidenced by the diplomatic fallout among Five Eyes allies.

What Deadlines Are Subscribers Facing?

The shutdown created financial consequences for paying subscribers. Today is the last day for subscribers who signed up between June 9 and June 14, the window spanning Fable 5's launch and the three days after the ban, to claim a prorated refund from Anthropic. Subscribers on Pro, Max, or Team plans who paid expecting access to Fable 5 but received effectively zero usable access must submit their refund request before the window closes.

A second deadline falls on June 23. That date was originally scheduled as the last day Fable 5 would be included at no additional cost within subscription plan limits. On June 23, Anthropic was scheduled to move Fable 5 to usage credits, a prepaid overage system billed at API rates of $10 per million input tokens and $50 per million output tokens, exactly double the cost of Claude Opus 4.8. Anthropic has issued no guidance on how the June 22 transition will be handled given the ongoing suspension.

The export control directive remains legally in effect. Restoring access would require either a formal Commerce Department withdrawal or a new authorization framework. No such framework has been announced, and no resolution timeline has been provided.