US Government Orders Anthropic to Suspend Fable 5 and Mythos 5 Globally Under Export Controls

What Happened

On June 12, 2026, at 5:21pm ET, the US government delivered an export control directive to Anthropic ordering the immediate suspension of access to its two flagship models, Fable 5 and Mythos 5, for all foreign nationals — including foreign national employees working at Anthropic itself inside the United States. Because enforcing that restriction selectively across a global customer base proved operationally impractical, Anthropic made the decision to disable both models for every customer worldwide rather than risk non-compliance.

Older Claude models are not affected by the directive. The suspension applies exclusively to Fable 5 and Mythos 5.

Why the Government Acted

The trigger was the discovery of a potential jailbreak technique targeting Fable 5. The method involved prompting the model to read a codebase and identify software flaws — a task that, under specific conditions, could reportedly be manipulated to elicit outputs the government considered sensitive enough to warrant export restrictions.

After reviewing the technique internally, Anthropic characterized it as consisting of minor vulnerabilities that are narrow in scope and non-universal. The company noted that the same outputs can be produced by other publicly available models without requiring any bypass at all, specifically citing GPT-5.5 (OpenAI) and its documented behavior at deploymentsafety.openai.com/gpt-5-5/cybersecurity as a direct comparator.

Anthropic's Response

Anthropic confirmed it is complying with the directive while making its disagreement explicit. In a public statement, the company said it "disagrees 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 pointed to the extensive pre-launch safety work it had completed before Fable 5 and Mythos 5 shipped, including red-teaming conducted jointly with the US government, the UK AI Safety Institute (AISI), and multiple private organizations. Anthropic had also built into Fable 5 a mandatory 30-day customer data retention requirement specifically designed to detect and respond to jailbreak attempts — a safeguard that, the company argues, demonstrates it was already treating the risk seriously.

Anthropic's core objection is proportionality: a narrow, non-universal technique that other frontier models can replicate does not, in its view, justify a global commercial suspension. The company did not indicate a timeline for restoration of access.

The Industry Precedent

The most consequential part of Anthropic's statement may be this line: "If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers."

That framing turns a single enforcement action into a policy flashpoint. If any narrow jailbreak — one that other models can also produce — is sufficient grounds for export-control suspension, then:

• No frontier model is likely to be fully immune, given the state of current red-teaming research.
• The practical effect could be a veto over commercial AI deployment wielded through export law rather than through sector-specific AI regulation.
• Foreign competitors operating outside US jurisdiction face no equivalent constraint, raising competitiveness concerns that are already central to the broader AI policy debate in Washington.

The Fable 5 / Mythos 5 suspension is the first known instance of the US government using export controls to force a domestic AI lab to pull a deployed consumer model. It arrives at a moment when Congress has yet to pass comprehensive AI legislation, leaving export control authorities as one of the few existing legal levers the executive branch can apply to frontier AI systems.

What Comes Next

Anthropic has published its account of the situation at anthropic.com/news/fable-mythos-access, a level of transparency unusual in government-directed compliance scenarios. The company's decision to publicly contest the directive's rationale, while simultaneously complying, signals it intends to fight the policy on its merits — likely through both public advocacy and direct engagement with regulators.

For the industry, the immediate question is whether this action represents a one-time response to a specific finding or the opening move of a broader export-control framework for advanced AI models. The answer will shape how every frontier lab thinks about model deployment, red-teaming disclosure, and the regulatory surface area of their most capable systems.