U.S. Government Forces Anthropic to Pull Fable 5 and Mythos 5 Offline Worldwide

By Hector Herrera | June 13, 2026 | Government · Breaking News

The U.S. Commerce Department ordered Anthropic to suspend all access to its Claude Fable 5 and Mythos 5 models for any foreign national — whether located inside or outside the United States — following a report that a company claimed to have jailbroken Mythos 5. Rather than attempt selective enforcement, Anthropic shut both models down entirely, cutting off customers globally just days after launch.

What Happened

The directive came from the Commerce Department's Bureau of Industry and Security (BIS), the agency that administers U.S. export controls on technology. The trigger was a claimed jailbreak of Mythos 5 — Anthropic's most capable frontier model — that alarmed administration officials over potential national security risks. Commerce responded by prohibiting Anthropic from making either Fable 5 or Mythos 5 accessible to any foreign national, a category broad enough to include Anthropic's own foreign-born employees working in the United States.

Anthropic's options were stark: implement identity-based access controls capable of screening every user and employee by citizenship status, or pull the models offline entirely. The company chose a full shutdown.

The Scope of the Order

The order applies to both models simultaneously. Fable 5 is Anthropic's mid-tier frontier model; Mythos 5 is its most capable, positioned against GPT-5.5 and Gemini Ultra. Both had launched within days of this directive.

Key facts:

• Agency: U.S. Department of Commerce, Bureau of Industry and Security
• Models affected: Claude Fable 5 and Mythos 5
• Scope: All foreign nationals — including Anthropic's own employees — inside and outside the U.S.
• Trigger: A company's claim to have jailbroken Mythos 5
• Anthropic's response: Full global shutdown of both models

The action represents one of the most direct applications of U.S. export control authority to a commercially deployed AI model. Export controls have historically governed hardware (like Nvidia GPUs) and software with specific dual-use applications, but applying them to access to a large language model's API is a significant extension of that authority.

Why This Matters

This is the first time a U.S. government agency has forced a major AI lab to pull a deployed frontier model offline on national security grounds. The precedent has immediate implications for the entire frontier AI industry.

Every lab that operates at the frontier — OpenAI, Google DeepMind, Meta — now knows that Commerce can issue a similar directive against their models if a jailbreak claim reaches the right desk at the right moment. The standard for triggering such an order appears to be a claimed jailbreak, not a confirmed one.

For Anthropic specifically, the shutdown disrupts customers who integrated Fable 5 or Mythos 5 into production pipelines. Enterprises and developers relying on those models now need to fall back to Claude Sonnet 4.6 or earlier versions — if those remain accessible — or migrate to competing providers.

The foreign national scope also creates an HR crisis within Anthropic itself. The company employs a large proportion of researchers born outside the United States, a pattern common across frontier AI labs. If those employees cannot access the company's most capable models, the product and research pipeline is directly affected.

The Jailbreak Trigger

The specifics of the claimed jailbreak have not been made public. It is not clear whether the company that claimed to have jailbroken Mythos 5 shared technical details with the Commerce Department or merely notified them of the claim. Jailbreaks — techniques that bypass a model's safety constraints — vary enormously in severity, from generating mildly policy-violating text to, in extreme cases, eliciting information with weapons-relevant applications.

The administration's decision to treat the claim as sufficient grounds for an export control order signals a risk tolerance much lower than what AI labs have operated under to date.

What to Watch

Google will appeal the order, and it's likely Anthropic is exploring legal options as well. The central legal question — whether a cloud API that serves responses constitutes an "export" under BIS regulations — has never been definitively adjudicated. Congress has been debating AI-specific export control legislation for two years; this action may accelerate that timeline.

Watch for other labs to proactively announce citizenship screening or geo-fencing measures as a hedge against a similar directive. That kind of preemptive compliance will reshape who can use frontier AI and under what terms — a structural shift that the industry has not yet priced in.