Anthropic Introduces CJS Jailbreak Severity Scale with Amazon, Microsoft, and Google

What happened

On July 2, 2026, Anthropic published technical details about Fable 5's cybersecurity protections and proposed the CJS (Cybersecurity Jailbreak Severity) scale — a cross-industry standard for evaluating AI jailbreak severity, developed jointly with Amazon, Microsoft, Google, and Glasswing ecosystem partners.

Context and impact

No unified standard previously existed for comparing jailbreak severity across AI companies. The CJS scale enables coordinated industry response and prioritized remediation. Alongside Fable 5, Anthropic also launched a HackerOne program for reporting jailbreaks.

Details

• CJS-0: Informational (no actionable capability)
• CJS-1 through CJS-3: Moderate levels with exponential severity increases
• CJS-4: Critical (direct critical infrastructure impact)
• Four scoring axes: capability gain, breadth of capability, ease of weaponization, discoverability
• Tier taxonomy: Prohibited / High-risk dual-use / Low-risk dual-use / Benign use
• New HackerOne program for security researchers to report jailbreaks