Three days. That is how long Anthropic’s most powerful AI models stayed online before the US government ordered them taken down.
On June 12, 2026, at 5:21pm ET, Anthropic received an export control directive from the Commerce Department instructing the company to suspend all access to Fable 5 and Mythos 5 by any foreign national, anywhere in the world. Because complying with that order while selectively blocking foreign users was operationally impossibl, Anthropic took the only viable path it shut both models down for everyone.
Hundreds of millions of users lost access within hours. The incident is the most dramatic government intervention in commercial AI deployment to date, and it has immediate implications for every enterprise that runs AI in production.
What Are Fable 5 and Mythos 5?
Anthropic launched Fable 5 and Mythos 5 on June 9, 2026. Mythos is the underlying model, a frontier-class system designed to reason across complex domains. Fable 5 runs on top of Mythos with a layer of safeguards built to limit access to the model’s most sensitive capabilities, particularly in cybersecurity so you can’t just ask it to hack the world.
The launch was notable. Fable 5 outperformed competitors across several industry benchmarks, and Anthropic had spent weeks working with the US government, the UK AI Safety Institute, and multiple private red-teaming organizations to stress-test the model’s safeguards before release.
The Jailbreak That Triggered the Shutdown
The government told Anthropic it had become aware of a technique to bypass Fable 5’s safeguards. The specific method involved asking the model to review a codebase and identify flaws, which caused the model to surface a small number of previously known, minor software vulnerabilities.
Anthropic reviewed the same technique. Its assessment: the vulnerabilities were minor, already publicly known, and discoverable using other models that remain deployed today, including OpenAI’s GPT-5.5. The company described the finding as a narrow, non-universal jailbreak, and noted that no testers had found a universal jailbreak that broadly bypasses safeguards across a wide range of harmful capabilities.
A narrow jailbreak produces some restricted output in a specific, constrained context. A universal jailbreak opens the model’s full restricted capability set to anyone who knows the technique. Anthropic had stated at launch that perfect jailbreak resistance is not achievable for any current model, and that the industry relies on defense-in-depth strategies rather than perfect safeguards. (This is why you need to Build Your Mythos Ready Security Program)
Why “Deemed Export” Rules Created a Global Off Switch
The legal mechanism behind the shutdown is the deemed export doctrine. Under US export control law, showing controlled technology to a foreign national inside the United States counts as an export to that person’s home country. For Fable 5, the government’s directive meant that any model output shown to a foreign national qualified as a controlled export.
The practical consequence was Anthropic could not comply by geo-blocking access in certain countries. Any foreign national employee at an American company, any international student at a US university, any non-citizen using the API from a US server would trigger a deemed export. The only compliant option was a global shutdown.
This is the first time deemed export rules have been applied to a commercial AI model at this scale, and it sets a precedent that cloud security and compliance teams need to understand.
Amazon’s Role in the Shutdown
The shutdown was not entirely government-initiated. Reports indicate Amazon CEO Andy Jassy brought the jailbreak finding to senior Trump administration officials before the directive was issued. Amazon is both a major investor in Anthropic and the cloud infrastructure provider that runs significant portions of Anthropic’s API.
That dual role creates a structural conflict of interest worth noting. Amazon had early access to information about the jailbreak, flagged it to the government, and its cloud infrastructure was the mechanism through which access was then cut. Enterprise customers should factor that relationship into their vendor risk assessments.
What Anthropic Says
Anthropic complied with the directive but rejected its basis publicly.
“We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people,” the company stated. “If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”
Anthropic also pointed out that the jailbreak demonstrated capabilities used every day by defenders who keep systems safe, and that the same capabilities are present in GPT-5.5, which remains deployed. The company called for a government process to review AI deployments that is “transparent, fair, clear, and grounded in technical facts.”
Anthropic says it is working to restore access and believes the shutdown represents a misunderstanding rather than a legitimate security threat.
What This Means for Cloud Security and Enterprise AI
This incident raises four concrete questions for cloud security teams.
1. Your continuity plans need to cover AI model availability
If a model your organization depends on can be pulled globally in under 24 hours without warning, your AI-dependent workflows need documented fallback procedures. The Fable 5 shutdown happened without advance notice, and most enterprise customers had no contingency in place.
2. Vendor concentration risk just got more visible
Anthropic, Amazon, and the US government are now entangled in ways that directly affect product availability. A model vendor, its cloud host, and its largest investor are effectively the same entity. That concentration creates a single point of failure that a traditional multi-vendor AI strategy does not fully address.
3. Deemed export compliance now applies to AI outputs
The deemed export doctrine has now been applied to AI model outputs in a way regulators will enforce. If your organization uses frontier AI in a context where foreign nationals access outputs, you may need to revisit your export compliance posture. The legal framework governing this is new and unsettled, but regulators have shown they will act on it.
4. The jailbreak standard for government action is unclear
The government acted on a narrow, non-universal jailbreak that Anthropic argued poses no meaningful uplift over existing deployed models. If that standard holds and expands to other models, it will materially affect how frontier AI products are deployed in regulated industries. Security teams should track how this standard evolves.
For now, all other Anthropic models, including Claude Opus, Sonnet, and Haiku, remain fully available. The directive applied only to Fable 5 and Mythos 5.
FAQ
What is Fable 5?
Fable 5 is Anthropic’s most capable commercial AI model, launched June 9, 2026. It runs on the Mythos underlying model and includes safeguards designed to limit access to sensitive cybersecurity capabilities.
Why did the US government shut it down?
The government cited national security concerns related to a jailbreak technique that allowed Fable 5 to surface software vulnerabilities when asked to review code. The directive was issued under export control authority.
What is the “deemed export” rule?
The deemed export doctrine treats showing controlled technology to a foreign national inside the US as an export to that person’s home country. The government applied this rule to Fable 5 model outputs, which required a global shutdown rather than a regional one.
Are other Claude models still available?
Yes. The directive applies only to Fable 5 and Mythos 5. All other Claude models, including Opus, Sonnet, and Haiku, remain accessible.
What should enterprises do now?
Document fallback procedures for AI-dependent workflows, assess your vendor concentration risk, and review your export compliance posture if your AI usage involves foreign national access to model outputs.
Will Fable 5 come back?
Anthropic says it is working to restore access and believes the shutdown is a misunderstanding. Restoration depends on resolving the government’s concerns about the jailbreak finding.
