If you use Claude, Anthropic's AI assistant, you may have noticed the most powerful version of it disappeared for about three weeks. Claude Fable 5 — the flagship model that competes with GPT-5 and Gemini's most capable tier — went offline globally on June 12 after the US government applied export controls (rules limiting which countries can access a technology) to it.

As of July 1, Fable 5 is back. But the path from ban to return has two parts that matter: what the US government extracted as conditions for lifting the ban, and what Anthropic is adding to its privacy policy on July 8.

Here's what happened, and what it means for you.

The Fable 5 ban arc

1. Jun 12

   Export controls imposed

   US Commerce Dept applies export controls after Amazon researchers demonstrate vulnerability exploit code generation from Fable 5 prompts. Global access suspended immediately for all users and API.

2. Jun 27

   Mythos 5 partial restoration

   Critical infrastructure organizations regain limited access to Mythos 5. Fable 5 remains offline globally. Face-scan biometric verification path announced for broader access.

3. Jun 30

   Export controls fully lifted

   Commerce Secretary Lutnick announces controls removed. Conditions: proactive security detection, malicious-activity reporting, co-development of future model standards with the government.

4. Jul 1

   Global access restored

   Fable 5 and Mythos 5 live on Claude.ai, Claude Code, and the developer API. 50% weekly usage cap in effect through July 7. Full uncapped access from July 7.

What triggered the ban

Amazon researchers found that Fable 5, under specific prompting patterns, would generate working code that demonstrated how to exploit software vulnerabilities. Not just describe the vulnerability in the abstract. Actual exploit code.

That's the kind of finding that gets a government's attention fast. The US Commerce Department applied export controls, which meant Anthropic couldn't keep Fable 5 available globally without verifying who was using it. Since verifying 100 million users' nationalities in real time was impossible, they took it offline.

The ban applied to everyone — paid subscribers, developers building apps on top of it, enterprise customers. Nineteen days without the most capable model.

What Anthropic agreed to in exchange

Commerce Secretary Howard Lutnick announced that Anthropic no longer required an export license after agreeing to three things: proactively detect and address security risks in Fable 5 and Mythos 5, work with the government on safety protocols for future model releases, and report any malicious activity they observe.

These are ongoing commitments, not a one-time fix. Anthropic is now formally engaged in reporting potential misuse of its own model to the US government, and in shaping the rules that will govern its next frontier releases.

Source spread

• CNBC — Trump admin lifts export controls on Fable 5 and Mythos 5 — builder. Lutnick quote and the three conditions, clearly stated.
• The Hacker News — Anthropic restores Fable 5 after jailbreak-linked controls — safety. Best coverage of the Amazon researchers' specific finding and why it triggered action.
• Forbes — Trump administration lifts export controls — builder. Clean summary of restoration mechanics and the 50% cap timeline.
• The Next Web — Export control lift analysis — skeptic. Notes the conditions are vague enough to function as a standing lever for future intervention.

Pros & cons

What's real:

• Fable 5 is back and available on Claude.ai for Pro, Max, Team, and Enterprise users. Free tier has limited access, subject to the 50% cap through July 7.
• The trigger for the ban was a genuine security concern, not politics. Exploit code generation from a publicly accessible frontier model is a real problem worth acting on.
• The conditions Anthropic agreed to are substantive: ongoing security detection, malicious-activity reporting, and co-developing safety standards with the government. This isn't a handshake deal.
• The biometrics requirement (effective July 8) applies to claude.ai users who want access to Fable 5, not to general Claude usage. Using Claude for everyday tasks doesn't require a face scan.

What deserves a side-eye:

• "Proactively detect and address security risks" is a commitment sentence with a lot of work to do. The public description gives the government ongoing reporting rights without specifying what threshold triggers a future access suspension. That's a standing lever, and it's deliberately vague.
• Anthropic updated its privacy policy to add biometrics collection (effective July 8) while most attention was on the export-control story. Government-issued ID and facial geometry templates (processed by a third-party vendor called Persona) are now part of the infrastructure. The company needs this to prove users are eligible for Fable 5 access.
• We haven't seen the full text of what Anthropic committed to. The description comes through a Commerce Dept press release and Anthropic's own statement. There may be more specifics that haven't been disclosed publicly.
• This ban is now the reference case for a playbook: suspend a frontier model, wait for the lab to negotiate under pressure, extract commitments. Nineteen days, done. The playbook works.

Anthropic no longer required an export license after agreeing to proactively detect and address security risks.

What to do about it

For everyday Claude users:

• If you use Claude.ai on the free tier: Fable 5 access is limited. You may encounter it in the model selector but usage will be capped at 50% of weekly limits through July 7, then available based on usage credits.
• If you're on Claude Pro, Max, or Team: Fable 5 is available now. The 50% weekly cap lifts July 7.
• About the biometrics policy (effective July 8): Accessing Fable 5 on Claude.ai after July 8 will require verifying your identity with a government-issued ID and a facial scan (processed by a company called Persona). General Claude usage for everyday tasks — asking questions, writing help, research — does not require this. Only Fable 5 access specifically.
• If you're not comfortable with the biometrics requirement: The other Claude models (Sonnet 5, Opus 4.8, Haiku) don't require it. Fable 5 is genuinely more capable, but the other models handle most everyday tasks fine. You can skip the face scan and still have a capable AI.
• Watch the privacy policy: Anthropic's updated policy takes effect July 8. Worth reading the Verification Data section specifically.

For builders

• Fable 5 API: claude-fable-5 is live. Full uncapped access from July 7.
• Mythos 5 API: claude-mythos-5 also restored on the same timeline.
• Biometrics requirement applies to claude.ai consumer accounts, not API. API builder access is not subject to ID verification.
• Update your service-dependency risk register: "Frontier model access suspension" is now a documented case with a documented timeline. 19 days, global scope, resolved via compliance negotiation. Not a theoretical risk.
• If you've built Fable 5 into a production product: document your fallback now. The June 12 ban is the reference case for the next one, whenever it comes.

Further reading

• Anthropic — Redeploying Claude Fable 5 — official restoration announcement
• CNBC — Trump admin lifts export controls on Fable 5 and Mythos 5 — Lutnick conditions and primary quote source
• The Hacker News — Security backstory on the Fable 5 ban — Amazon researchers' finding and technical backstory
• Forbes — Full restoration details — 50% cap mechanics and timeline
• The Next Web — Export control lift analysis — skeptical read on conditions' vagueness
• Anthropic — Privacy policy — biometrics addition effective July 8