Analysis · curated 11 Sep 2026
Why LLM jailbreak classification doesn’t work like CWE
First reported gopenai.com
Coverage timeline
Single-source analysis — first reported, latest, and curated coincide.
Why it matters
Jailbreak taxonomy discussions help defenders understand why traditional vulnerability-enumeration approaches fail for LLMs and why attack-surface coverage for models requires different frameworks.
An analysis by Ivan Ponurovskiy argues that LLM jailbreaks cannot be classified like CWE-style software weakness enumerations because jailbreaks are heuristic, arising from the probabilistic nature of models and alignment rather than deterministic flaws. The piece references a domain-based taxonomy of jailbreak vulnerabilities (mismatched generalization, competing objectives, adversarial robustness, and mixed attacks) rather than presenting new experimental findings.