Anthropic Disables Claude Fable 5 and Mythos 5 After US Government Order

Anthropic has disabled two of its most capable models for each customer. The shutdown followed a US government order to control exports. The order arrived on June 12, 2026. Named Claude Fable 5 and Claude Mythos 5 directly. Both models were launched just three days earlier, on June 9.

The order cited national security authorities, according to Anthropic. Stop access to any foreign person, inside or outside the United States. That estimate included Anthropic's foreign workers. Anthropic cannot filter out foreigners from US users in real time. So shut down both models for everyone to ensure compliance.

Access to all other Anthropic models was not affected. Claude Opus 4.8 and others live online.

So, What Actually Happened

Anthropic published a public statement within hours of ordering. Commerce Secretary Howard Lutnick sent a letter to CEO Dario Amodei. The letter did not specifically address national security concerns.

Anthropic's own learning is small. He believes the government has learned how to “break the prison” Myth 5. Jailbreaking bypasses model protections to extract blocked behavior. Anthropic reviewed the demonstration related to less vulnerability. It says that other public models get the same errors without passing. Anthropic also said the power shown is widely available elsewhere. It named OpenAI's GPT-5.5 as a comparable example.

Why the Government took action

The Commerce Department moved after another company claimed to have hacked Mythos. That claim alarmed officials about the potential threat to national security. Earlier, the administration tried to postpone the launch. Anthropic refused, and an export control order followed.

Anthropic is subservient while clearly opposing reason. Call this situation a possible misunderstanding. The company says that the jailbreaks identified are few and far between. The plan essentially asks the model to read the codebase and fix the errors. Anthropic says a minor jailbreak shouldn't justify a full recall. It warns that the rate will freeze new shipments across the industry.

Which Models Are Affected

The table below summarizes the current status of each model.

How Fable 5's Safeguards Work

Fable 5 is a Mythos-class model, a tier above Opus. Anthropic has made it safe for general use with new classifiers. Classifiers are different AI systems that detect potential abuse. They flag questions about cybersecurity, biology and chemistry, and distillation. Flagged queries revert to Claude Opus 4.8 instead. Users are notified whenever an error occurs.

According to Anthropic, relapse starts in less than 5% of sessions. Over 95% of fantasy moments don't involve going back at all. The company has also adopted a 'defense in depth' strategy. It aimed to make jailbreaks smaller or more expensive to produce. Match this with 30-day data retention to detect novel attacks. That data is not used to train new models.

The Anthropic red-team put together these defenses before launch thousands of hours. Our partners include the US government, UK AISI, and foreign organizations. An external bug bounty has not found a worldwide jailbreak in over 1,000 hours. The universal jailbreak bypasses defenses in all multiple abilities. Anthropic also said no testers have found one in production yet. It admits that complete jailbreak resistance is almost impossible for anyone.

Use Cases Now Interrupted

Early adopters conducted their own tests before closing. Stripe said Fable 5 compressed months of engineering into days. Implemented extensive codebase migration to the 50 million line Ruby codebase. Financially, Hebbia ranked the Fable 5 the highest on its bench. In theory, Fable 5 rebuilt the web application code from screenshots. It also beat Pokémon FireRed with only a vision harness.

Mythos 5 conducted a scientific mission with Project Glasswing. In-house experts have accelerated the composition of the drug by about ten times. Glasswing partners use Mythos class models to debug software. Mozilla alone has reported solving hundreds of vulnerabilities this way. That protective workflow is now paused by shutdown.

Mythos 5 & Mythos 5 Shutdown Explorer

Migrate Your Code Today

The developers called Fable 5 the Claude API. The string was the model claude-fable-5. Requests to it now fail while the suspension is in place. The pattern below dates back to Opus 4.8 graciously.

import anthropic

client = anthropic.Anthropic()

def ask(prompt, primary="claude-fable-5", fallback="claude-opus-4-8"):
    for model in (primary, fallback):
        try:
            return client.messages.create(
                model=model,
                max_tokens=1024,
                messages=[{"role": "user", "content": prompt}],
            )
        except anthropic.APIStatusError as err:
            # Fable 5 is currently disabled; route to Opus 4.8.
            print(f"{model} unavailable: {err.status_code}")
    raise RuntimeError("No model available")

print(ask("Summarize this change log.").content[0].text)

Key Takeaways

• A US export control order forced Anthropic to disable Fable 5 and Mythos 5 on June 12, 2026.
• All other Claude models, including the Opus 4.8, remain fully stocked.
• The trigger was a claim by another company to jailbreak; Anthropic calls it small and non-ubiquitous.
• Fable 5's classifiers route cyber, bio-chem, and distillation questions in Opus 4.8 less than 5% of the time.
• This appears to be the first government-enforced reduction of the publicly invested border model.

How Developers Respond to Shutdowns

Sources