jailbreak-taxonomy

A jailbreak taxonomy catalogs prompt injection and adversarial robustness mechanism families, mapping each to actionable defenses. It provides structured detection signatures and severity ratings for red-team testing, policy hardening, and risk assessment across LLM deployments.

Map prompt injection threats to defenses by cataloging jailbreak techniques across six mechanism families. The taxonomy specifies technique details, detection signatures, and severity, enabling structured defense mapping and escalation guidance for security teams.

Yes, the taxonomy supports system prompt hardening and risk assessment across model prompts and outputs. It applies structured defense mapping to identify vulnerabilities and specify actionable defenses against adversarial robustness threats.

The taxonomy covers six main jailbreak mechanism families. It categorizes adversarial robustness techniques by specifying their technique details, detection signatures, and severity to assist in comprehensive threat modeling and response.

Yes, the taxonomy provides detection signatures and severity levels for each jailbreak mechanism family. This structured format directly supports incident response by offering defense mapping and escalation guidance for compromised AI safety systems.

Structure threat model defenses by mapping six jailbreak mechanism families to actionable defenses. The taxonomy outputs technique details, detection signatures, and severity ratings, ensuring comprehensive risk assessment and policy hardening.