Anthropic Pulls Claude Fable 5 Offline Again After Security Researcher Pliny Reports Successful Jailbreak

Claude Fable 5 has lived through one of the roughest launch cycles in AI history. Anthropic released the Mythos-class model on June 9, 2026, only to see researcher Pliny the Liberator report a jailbreak within 48 hours. The fallout triggered an export control shutdown on June 12. Today, July 1, Anthropic restores Claude Fable 5 with tighter classifiers, closing a three-week saga that rattled AI investors and partners including Amazon (AMZN), Microsoft (MSFT), and Alphabet (GOOGL).

The Jailbreak That Started It All

Pliny the Liberator announced his bypass just 48 to 72 hours after launch. He used Unicode homoglyphs, narrative framing, and decomposition tactics to evade Fable 5’s classifiers.

The researcher also leaked Fable 5’s full 120,000-character system prompt to GitHub’s CL4R1T4S repository. Anthropic had claimed 1,000-plus bug bounty hours produced zero universal jailbreaks before release.

Amazon’s Role in the Shutdown

• Amazon researchers reportedly triggered the export control review.
• The jailbreak allegedly flagged software flaws and generated exploit code.
• Anthropic disputed the severity, noting similar prompts work on Opus 4.8 and rival models.

That finding pushed regulators to act fast, cutting access within hours of the report.

Claude Fable 5 will be available again globally tomorrow.

After a series of productive conversations with the US government, we're redeploying the model with a new set of classifiers to target and block more cybersecurity tasks. In the near term, some routine tasks like coding…

— Anthropic (@AnthropicAI) July 1, 2026

Export Controls Force a Global Blackout

On June 12, 2026, a U.S. export control order barred foreign nationals from using Claude Fable 5 or Mythos 5. Anthropic couldn’t verify nationality in real time.

The company shut both models down worldwide rather than risk violations. Partial access returned June 26 for roughly 100 U.S. companies and federal agencies protecting critical infrastructure.

Claude Fable 5 Returns Today

Commerce Secretary Howard Lutnick approved renewed access after a two-week review with Anthropic. Claude Fable 5 relaunches today across Claude.ai, Claude Platform, Claude Code, and Claude Cowork.

New Classifier Cuts Repeat Jailbreaks

Anthropic’s new classifier reportedly blocks the leaked technique in over 99% of attempts, per its June 30 update. Flagged coding requests now fall back to Opus 4.8 instead of degrading silently.

• Classifier success rate: 99%+ against the leaked technique
• Fallback model: Claude Opus 4.8
• Update published: June 30, 2026

Bounty Program and Industry Framework

Anthropic also opened a HackerOne bounty program for future Fable 5 jailbreak reports. It’s drafting a shared severity framework with Amazon, Microsoft, and Google under its Glasswing partnership. That framework aims to standardize how AI developers rate and respond to jailbreak risks going forward.

Disclaimer:

The content shared by Meyka AI PTY LTD is solely for research and informational purposes. Meyka is not a financial advisory service, and the information provided should not be considered investment or trading advice