Anthropic Claude Fable 5 has been blocked — and it happened unexpectedly quickly. On June 12, 2026, exactly three days after the model’s public release, the U.S. Department of Commerce issued a directive for the immediate shutdown of Claude Fable 5 and Claude Mythos 5. The company complied with the order within a few hours. The reason was a discovered method for bypassing the model’s safety filters. But Anthropic strongly disagrees with the decision, and this dispute is already turning into one of the biggest regulatory conflicts in the history of the AI industry.
Timeline: How It All Happened in 72 Hours
To understand the scale of the event, it is important to see the full picture over time.
- June 9, 2026, 00:00 — Anthropic publicly releases Claude Fable 5 — the first public Mythos-class model. On the same day, CEO Dario Amodei publishes the article “Policy on the AI Exponential,” warning about the risks of frontier AI and asking governments to activate regulatory mechanisms
- June 10-11 — Fable 5 is available to millions of users through Claude.ai Pro, Max, Team and Enterprise subscriptions, as well as the API
- June 12, 17:21 ET — Anthropic receives a directive from the U.S. Department of Commerce, signed by Secretary Howard Lutnick. The letter is sent directly to CEO Dario Amodei
- June 12, evening — Anthropic disables both models globally for all users and publishes an official statement
- June 13 — CNN, WSJ, TechCrunch and NBC News publish reports. The story reaches the front pages of technology media
Three days from a triumphant release to shutdown. This is a record-short active lifespan for a public frontier model.
What Exactly the Government Blocked and Why
The U.S. Department of Commerce directive prohibited access to Claude Fable 5 and Claude Mythos 5 for any foreign national — regardless of whether they are located inside the United States or abroad. The formal basis was an export-control directive citing national security concerns.
This detail is critically important: the wording “any foreign national” meant that even Anthropic’s own employees with foreign passports were not allowed to use the models. Anthropic technically cannot filter U.S. citizens from foreign nationals in real time across its user base. The only solution that made it possible to comply with the directive was to shut down both models for everyone.
What a Jailbreak Is and Why the Government Reacted This Way
Anthropic said it received only a verbal warning about a “narrow, non-universal jailbreak” of Fable 5. According to the company’s understanding, the issue was this: someone found a way to bypass Fable 5’s safety classifiers through a request asking the model to analyze code and find vulnerabilities in it.
This capability — cybersecurity tasks — is precisely the area that Anthropic deliberately blocked through its filter system at launch. As we described in the article about the Fable 5 launch, the model switches cybersecurity-related requests to the older Opus 4.8. But it turned out that this protection could be bypassed.
Anthropic’s Position: “This Is the Wrong Standard”
Anthropic did not stay silent. The company published an official statement and openly disagreed with the government’s decision: “We disagree that the discovery of a narrow potential jailbreak should be grounds for withdrawing a commercial model deployed for hundreds of millions of people.”
Anthropic then issued a broader warning: “If this standard were applied across the industry, we believe it would effectively halt all new model deployments by all frontier model providers.”
The company also emphasized that the discovered vulnerabilities are simple, already known and can be found by other publicly available models — including GPT-5.5. In other words, blocking Fable 5 specifically does not solve the problem; it only demonstrates the government’s power.
The Irony of the Situation: Anthropic Asked for Regulation and Got It
This is probably the worst possible scenario for Anthropic from an image perspective.
For years, the company positioned itself as the “safe alternative” in the AI industry. It was Anthropic that publicly called on governments in different countries to introduce regulatory mechanisms for controlling frontier models. It was CEO Dario Amodei who — two days before the block — published a piece insisting on the need for government intervention in AI development and cited Mythos as an “illustrative example” of the danger posed by powerful models.
The same administration that tried to reduce oversight mechanisms in January 2025 is now, in June 2026, hesitantly using increasingly harsh levers of state power to control the AI industry — and Anthropic in particular.
Anthropic asked for regulation — and the government used it to shut down Anthropic itself. This is not malicious intent; it is the logical consequence of what happens when regulatory mechanisms exist without a clear legal framework.
Context: Anthropic’s Tense Relationship With Washington
The Fable 5 shutdown is not an isolated event, but part of a prolonged conflict between Anthropic and the Trump administration. Here are the key moments:
- Early 2025 — the Trump administration cancels Biden-era AI oversight mechanisms; Anthropic publicly criticizes the decision
- February 2026 — conflict between Anthropic and the Pentagon over the terms of using Mythos for military tasks
- April 2026 — launch of Project Glasswing; regulators around the world raise alarms over Mythos capabilities (we covered this in a previous article)
- June 9, 2026 — launch of Fable 5; Amodei publishes a warning about AI risks
- June 12, 2026 — the government shuts down the models
In other words, the government effectively used Anthropic’s own public warnings as justification for action. This is an extremely uncomfortable position for the company.
What Remains Online and What It Means for Users
An important practical detail: the directive applies only to Fable 5 and Mythos 5. All other Anthropic products remain fully accessible.
What works without changes:
- Claude Opus 4.8 — the previous flagship, available through Claude.ai and the API
- Claude Sonnet 4.6 — the standard working model
- Claude Haiku 4.5 — the fast and lightweight version
- Anthropic web crawlers (ClaudeBot, Claude-User, Claude-SearchBot) — continue operating
For most users and businesses, the practical impact is minimal: Fable 5 was available for only 3 days. Most enterprise and API clients continue working on Opus 4.8 without changes.
What Happens Next: IPO, Court and an Industry Precedent
This situation has several parallel consequences that will unfold over the coming weeks and months.
Anthropic’s IPO is at risk. The company is preparing for a public offering expected in October 2026. The shutdown of its most powerful model three days after release is a serious negative signal for potential investors. Questions of trust in product releases and regulatory risks will be at the center of any roadshow.
Legal conflict. Anthropic said it is “working to restore access” and is challenging the decision. This almost certainly means a legal process. The government has not yet provided detailed technical evidence of the jailbreak — only a verbal notice.
A precedent for the entire industry. This appears to be the first forced government shutdown of a publicly deployed frontier model. If the U.S. government can shut down any AI model from any company on the basis of a “narrow potential jailbreak,” this fundamentally changes the rules of the game for all AI labs. OpenAI, Google DeepMind and Meta AI are closely watching how Anthropic gets out of this situation.
Europe’s reaction. Reports are already appearing that the EU sees the Fable 5 situation as a “wake-up call” about dependence on American AI platforms and the need for its own regulatory architecture.
In Brief: The Key Facts About the Blocking of Anthropic Claude Fable 5
- Anthropic Claude Fable 5 was blocked on June 12, 2026, by order of the U.S. Department of Commerce
- The directive was signed by Secretary Howard Lutnick; the letter was addressed to CEO Dario Amodei
- Basis: export-control directive banning access for foreign nationals
- Reason, according to Anthropic: a discovered narrow jailbreak via code analysis
- Anthropic disagrees: the same method is available in GPT-5.5 and other models
- Disabled: Fable 5 and Mythos 5. Still online: Opus 4.8, Sonnet, Haiku
- At risk: Anthropic’s IPO, expected in October 2026
- The first forced rollback of a public frontier AI model in history
FAQ
Why did Anthropic shut down Claude Fable 5? Because of a mandatory directive from the U.S. Department of Commerce dated June 12, 2026, which prohibited foreign nationals from accessing the models. Since Anthropic cannot filter users by citizenship in real time, the company shut down both models for everyone.
Is Claude available now? Yes. Claude Opus 4.8, Sonnet 4.6 and Haiku 4.5 remain fully available. Only Fable 5 and Mythos 5 are blocked.
Does Anthropic agree with the government’s decision? No. The company is legally complying with the directive, but publicly disagrees and considers the discovered jailbreak “narrow and non-universal,” available in other public models as well.
The article was prepared by the TechVisor team — practical IT media for people.
