Anthropic Redeploys Fable 5 and Proposes a Cross-Lab Rubric for Scoring Jailbreaks

Anthropic restored global access to Claude Fable 5 on July 1 after export controls that had limited the model were lifted, and paired the return with a proposed industry standard for measuring jailbreak risk. The company published detailed documentation on the model's cybersecurity safeguards alongside the redeployment.

The main proposal is a Cyber Jailbreak Severity (CJS) framework, developed with Amazon, Microsoft, Google, and other partners under the Glasswing grouping. It scores a jailbreak on four criteria: capability gain, breadth of that gain, ease of weaponization, and discoverability. It then maps the summed score onto four bands from low to critical. The intent is a shared vocabulary so labs and governments can compare incidents rather than each grading its own.

Fable 5's safeguards also address the dual-use problem directly. Rather than blocking all security-related requests, its classifiers sort cyber queries into categories, and Anthropic opened a HackerOne bounty inviting researchers to find bypasses.

A severity rubric is a governance instrument, not a defense. It standardizes how a jailbreak is described after the fact and depends on labs reporting honestly. Whether competitors adopt CJS or treat it as one vendor's schema will decide if it becomes a real benchmark or only a marketing effort.