US Government Forces Fable 5 Offline: How One Jailbreak Demo Took Down Anthropic's Flagship Model

Claude Fable 5 and Claude Mythos 5 launched on June 9, 2026, making Mythos-class capabilities publicly available for the first time. Three days later, at 5:21 PM ET on June 12, Anthropic received an export control directive from US Commerce Secretary Howard Lutnick and was forced to shut down both models for all global users within hours.

This marks the first time a government has used national security authority to compel the shutdown of a publicly deployed frontier AI model.

Three Days from Launch to Shutdown

The public release of Fable 5 was itself a significant milestone. Anthropic packaged Mythos-level capabilities into a version with additional safety guardrails and opened it to the public, free for Pro and Max subscribers through June 22. Pricing was set at $10 per million input tokens and $50 per million output tokens, roughly an order of magnitude above Opus 4.8.

The day after launch, on June 10, a pseudonymous researcher known as “Pliny the Liberator” posted a series of jailbreak demonstrations on X. The techniques involved Unicode substitution, homoglyphs, and character decomposition to bypass Fable 5’s safety classifiers. The full 120,000-character system prompt was uploaded to GitHub, and the leaked outputs included stack buffer overflow exploitation guidance and methamphetamine synthesis steps.

Crucially, a separate undisclosed organization claimed to possess a similar jailbreak method and reported it to government officials, directly triggering the shutdown order.

The Directive and Anthropic’s Response

In its official statement, Anthropic described the Commerce Department directive as prohibiting “any foreign national from accessing Fable 5 and Mythos 5, whether inside or outside the United States, including Anthropic’s own foreign national employees.” To ensure compliance, the company’s only option was to disable access for all customers worldwide.

Anthropic explicitly disagreed with the decision. The company’s technical assessment found that the jailbreak method reproduced only “a small number of already known, minor vulnerabilities,” the same vulnerabilities present in competing models including GPT-5.5. Anthropic stated directly that applying this standard across the industry would “essentially halt all new model deployments.”

As of June 15, access to Fable 5 and Mythos 5 remains suspended. Anthropic says it is working to restore access but has not provided a timeline. All other Claude models, including Opus 4.8, Sonnet 4.6, and Haiku 4.5, continue to operate normally.

The “Hardware Sovereignty” Reaction

The shutdown triggered a notable shift in developer communities. Discussions urging teams to “run local models on home GPUs” went viral, with open-weight alternatives like Kimi K2.7 Code gaining sudden attention. The logic is straightforward: if your workflow depends on a cloud API, a single government directive can take your system offline in hours.

The “hardware sovereignty” framing had been circulating in niche circles for months. The Fable 5 incident moved it from theoretical concern to concrete evidence, particularly for enterprise teams that had just started onboarding Fable 5 for production use.

Anthropic’s own situation adds a layer of complexity. On June 3, the company had confidentially filed its S-1 with the SEC, targeting a fall 2026 IPO at a roughly $965 billion valuation. Fable 5 was a critical driver of new revenue growth. Every day offline carries real financial cost.

Where Regulatory Logic Breaks Down

What makes this incident worth watching is not just that Fable 5 was taken down, but the regulatory pattern it reveals.

Export controls were designed for hardware: chips, physical devices, things that can be tracked and restricted at borders. Applying that framework to a software model deployed via API creates a very different dynamic. The directive arrived without technical specifics, required global action within hours, and gave Anthropic no mechanism to dispute or delay. The company disagreed, complied anyway, and is now navigating a process with no defined timeline.

Digital Watch Observatory’s analysis notes a sharp contrast with the EU’s risk-based regulatory approach, which favors pre-deployment assessment and ongoing audits over emergency shutdowns.

Two scenarios are now in play. Anthropic patches quickly and restores access within days, turning this into a forced rehearsal for government-AI security coordination. Or the restoration drags, the negotiations go public, and the pressure for a formalized US AI regulatory framework intensifies considerably earlier than most in the industry expected.

If this was useful, subscribe to the newsletter for weekly AI PM insights and GenAI case studies.

---

Related reading:

• Anthropic Mythos: AI Finds Vulnerabilities Faster Than Humans Can Patch Them
• Anthropic Files Confidential S-1: The $965 Billion AI IPO