US Government Orders Anthropic to Suspend Claude Fable 5

Anthropic disabled Claude Fable 5 and Claude Mythos 5 on June 12, 2026 — just three days after the models launched — to comply with a US government export-control directive. The order, citing national security authorities, suspends access for any foreign national, and Anthropic says the practical result is a global shutdown of both models for all customers. The company is disputing the rationale while working to restore access.

What happened

Anthropic received the directive at 5:21pm ET on June 12, 2026. The government ordered the company to suspend all access to Fable 5 and Mythos 5 "by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees." With no way to verify citizenship per request in real time, Anthropic disabled both models entirely.

The timing made the disruption particularly striking. Anthropic had launched Claude Fable 5 on June 9 — the first publicly available Mythos-class model, priced at $10 per million input tokens and $50 per million output tokens. Claude Mythos 5, the unrestricted sibling, was already limited to vetted Project Glasswing partners. Both were offline within 72 hours.

Other Claude models — Opus 4.8, Sonnet 4.6, and Haiku — are unaffected and remain fully available.

The jailbreak dispute

The government's letter cited a method of bypassing, or jailbreaking, Fable 5's safety safeguards. Anthropic reviewed the demonstration and disputes the characterization.

In a public statement, the company describes the technique as "a narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws." Fortune later reported the prompt was simply "fix this code." Anthropic says the method surfaces "previously known, minor vulnerabilities" that are "relatively simple" and discoverable by other public models like GPT-5.5 — and that accepting this standard for recalling a model would "essentially halt all new model deployments for all frontier model providers."

Anthropic had red-teamed Fable 5's safeguards for thousands of hours with the US government, UK AISI, and third-party organizations before launch. No tester found a universal jailbreak, and the company has stated it believes perfect jailbreak resistance is not currently possible for any model provider.

Who triggered the directive

Reporting that emerged after the shutdown points to Amazon as a key actor. Citing sources, outlets including CNBC and Bloomberg reported that Amazon CEO Andy Jassy briefed Treasury Secretary Scott Bessent that Fable 5 could be used to extract cyberattack information. Amazon researchers had conducted the analysis that formed the basis of the government's concern. Anthropic and Amazon have a complex relationship — Amazon is Anthropic's largest cloud investor but also competes through AWS Bedrock and its own AI offerings.

Axios further reported that the administration alleges Anthropic allowed Mythos 5 to be accessed by an entity with ties to China, and that the company failed to comply with a recent cyber executive order. Anthropic has not publicly addressed those allegations.

What it means for developers

Anyone building on Fable 5 lost access to the model without notice. API calls to claude-fable-5 return errors, and the model selector on claude.ai no longer lists it. Applications that hardcoded the model ID need a code change to point at an alternative.

Claude Opus 4.8 is the immediate fallback within Anthropic's lineup — still a capable model for coding, writing, and analysis at half the price ($5/$25 per million tokens). Claude Code continues to work with Opus 4.8 and Sonnet 4.6. For teams that need capability at a similar level, our Claude Fable 5 review covers the full landscape of alternatives, and our best AI chatbots guide ranks the field.

The incident is also a practical lesson in model dependency. A model your product depends on can vanish within hours for reasons that have nothing to do with its technical quality. Treating the model ID as replaceable configuration — not a hardcoded constant — is now a production requirement, not an architectural preference.

What to watch next

Anthropic's senior technical staff were in Washington, D.C. as of June 15, meeting with White House and Commerce Department officials to dispute the ban. The company has not provided a restoration timeline. The outcome of those meetings — and whether the government accepts Anthropic's argument that the jailbreak is neither novel nor model-specific — will determine whether Fable 5 returns to the public or becomes a landmark case in how frontier AI is regulated.