Anthropic Details Claude Fable 5 Cybersecurity Safeguards and Jailbreak Framework

Anthropic published detailed technical documentation on Claude Fable 5’s cybersecurity safeguards, describing a four-category safety classifier (Prohibited, High-risk dual use, Low-risk dual use, Benign) and proposing a Cyber Jailbreak Severity (CJS) framework (CJS-0 to CJS-4) with four scoring axes to standardize ratings of jailbreak risk; the disclosure aims to balance dual-use concerns, request community feedback, and channel reports via a HackerOne program rather than reporting any active incident.