US Government Order Forces Anthropic to Disable Claude Fable 5 and Mythos 5

Anthropic has disabled its two most capable Claude models for all customers after receiving a US government export control directive, marking what appears to be the first government-forced takedown of a publicly deployed frontier model.

What Happened

On June 12, 2026, Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei naming Claude Fable 5 and Claude Mythos 5 specifically. The directive, citing national security authorities, suspended access by any foreign national—inside or outside the United States, and including Anthropic's own foreign national employees.

Because Anthropic cannot filter foreign nationals from US users in real time, it disabled both models entirely to ensure compliance. The shutdown landed just three days after the models launched on June 9. Access to all other Anthropic models, including Claude Opus 4.8, Sonnet, and Haiku, was unaffected.

According to Anthropic's public statement, the letter did not spell out the specific national security concern. The company's own reading is that the government learned of a method to "jailbreak" Fable 5. The Commerce Department reportedly moved after another company claimed it had jailbroken Mythos, a claim that alarmed officials. The administration had earlier tried to delay the launch; Anthropic declined, and the export control letter followed.

Anthropic is complying while disputing the rationale. It says the demonstrated jailbreak is narrow and non-universal—the technique essentially asks the model to read a codebase and fix flaws—and that other public models, including OpenAI's GPT-5.5, find the same flaws without any bypass. The company warns that treating a narrow jailbreak as grounds for a full recall would halt new deployments industry-wide.

Model Status

How Fable 5's Safeguards Worked

Fable 5 is a Mythos-class model, a tier above Opus. Anthropic made it safe for general use with classifiers—separate AI systems that detect potential misuse and flag queries on cybersecurity, biology and chemistry, and distillation. Flagged queries fall back to Claude Opus 4.8, and users are notified when a fallback occurs. Anthropic says fallback triggers in under 5% of sessions, with over 95% seeing none.

The company describes a "defense in depth" strategy paired with 30-day data retention to detect novel attacks; that data is not used to train new models. Anthropic says it red-teamed the safeguards for thousands of hours with partners including the US government, the UK AISI, and external organizations. An external bug bounty found no universal jailbreak in over 1,000 hours, and the company says no tester has found one on production tasks—while conceding perfect jailbreak resistance is likely impossible for anyone.

Disrupted Workflows

Early adopters had begun integrating both models. Stripe said Fable 5 ran a codebase-wide migration on a 50-million-line Ruby codebase, compressing months of engineering into days. Hebbia ranked it highest on its finance benchmark, and the model rebuilt a web app from screenshots and beat Pokémon FireRed with a vision-only harness. On the Mythos side, Project Glasswing partners used the models for defensive security work—Mozilla reported resolving hundreds of vulnerabilities—and internal experts cited roughly tenfold acceleration on parts of drug design. Those workflows are now paused.

Migrating Today

Developers called Fable 5 via the model string claude-fable-5, which now errors. Anthropic recommends Opus 4.8 as the fallback target. A graceful pattern:

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:
            print(f"{model} unavailable: {err.status_code}")
    raise RuntimeError("No model available")

Why It Matters

If the source reporting holds, this is the first time a publicly deployed frontier model has been pulled offline by government order. Anthropic says it is working to restore access quickly and calls the situation a likely misunderstanding. Regardless of how this specific dispute resolves, the episode establishes a precedent: national security authorities can be invoked to force a frontier model offline within hours, on a global basis, over a contested jailbreak claim. For teams building on frontier APIs, that introduces a new category of availability risk—one that fallback routing can mitigate operationally but not eliminate. This is a developing story.