“Anthropic Yanks Claude Fable 5 and Mythos 5 Following Federal Directive”

Anthropic has taken its two most powerful models offline for all users. The action came in response to a US government export control order. The directive was received on June 12, 2026. It singled out Claude Fable 5 and Claude Mythos 5 by name. Both had been released just three days prior, on June 9.

The order invoked national security grounds, according to Anthropic. It barred access for any foreign national, whether inside or outside the US borders. Even Anthropic’s own employees who are foreign nationals were caught by the restriction. The company lacks the ability to separate foreign nationals from domestic users in real time. To remain compliant, it pulled both models for every customer.

All other Anthropic models remained accessible. Claude Opus 4.8 and the remainder continued running normally.

So, What Actually Happened

Anthropic issued a public response within hours of receiving the directive. Commerce Secretary Howard Lutnick addressed the letter directly to CEO Dario Amodei. The correspondence did not detail the specific national security threat involved.

Anthropic’s internal assessment is more limited in scope. It suspects the government became aware of a technique to “jailbreak” Fable 5. A jailbreak circumvents a model’s safety controls to produce restricted outputs. Anthropic examined a demonstration linked to several minor vulnerabilities. It notes that other publicly available models share the same weaknesses without requiring any bypass. The company also pointed out that the demonstrated capability is readily accessible from other sources. It cited OpenAI’s GPT-5.5 as a comparable instance.

Why the Government Acted

The Commerce Department took action after a separate company reported jailbreaking Mythos. That report raised alarms within the administration about potential national security exposure. Officials had previously attempted to persuade Anthropic to delay the rollout. The company refused, and the export control directive followed as a consequence.

Anthropic is following the order while publicly challenging its reasoning. It characterizes the episode as most likely a misunderstanding. The firm argues the referenced jailbreak is narrow in scope and not broadly applicable. The technique basically instructs the model to analyze a codebase and patch defects. Anthropic contends that such a limited security gap should not warrant a complete recall. It cautions that applying this benchmark would effectively freeze new model launches across the entire industry.

Which Models are Affected

The following table outlines the present status of each model.

How Fable 5’s Safeguards Work

Fable 5 belongs to the Mythos class, a rank above Opus. Anthropic engineered it for general availability through additional classifiers. Classifiers are independent AI systems designed to spot possible abuse. They screen queries related to cybersecurity, biology, chemistry, and distillation attempts. Any flagged query gets redirected to Claude Opus 4.8 instead. Users receive a notification each time a redirection occurs.

Anthropic reports that redirections activate in fewer than 5% of sessions. More than 95% of Fable usage proceeds without any redirection whatsoever. The firm also employed a layered defense strategy. Its goal was to ensure jailbreaks remained narrow or prohibitively costly to engineer. This approach was combined with a 30-day data retention window to catch emerging attack methods. That retained data is not fed back into model training.

Before releasing Fable 5, Anthropic stress-tested these protections for thousands of hours. Testing partners included the US government, the UK AI Safety Institute, and outside organizations. An external bug bounty program failed to uncover any universal jailbreak after more than 1,000 hours of effort. A universal jailbreak would broadly overcome safeguards across multiple capability areas. Anthropic further confirmed that no tester has identified one on live production workloads so far. It acknowledges that flawless jailbreak resistance is most likely unattainable for any provider.

Use Cases Now Disrupted

Early customers conducted their own evaluations before the shutdown took effect. Stripe reported that Fable 5 condensed months of engineering work into mere days. It handled a full-codebase migration on a 50-million-line Ruby project. In financial analysis, Hebbia placed Fable 5 at the top of its benchmark rankings. On the vision front, Fable 5 reconstructed a web app’s source code from screenshots alone. It also completed Pokémon FireRed using nothing but a vision-based setup.

Mythos 5 powered scientific research through Project Glasswing. In-house specialists sped up portions of drug design by roughly tenfold. Glasswing partners leveraged Mythos-class models to patch software vulnerabilities. Mozilla alone confirmed it resolved hundreds of security issues using this workflow. Those defensive operations are now on hold because of the shutdown.

Fable 5 & Mythos 5 Shutdown Explorer