Research
SpecGuard: Detecting Multi-Turn LLM Jailbreaks via Specification-Based Runtime Monitoring
Publication date unknown · Discovered arxiv.org
Page published
Publication date unknown · First observed: 30 Oct 2026
Coverage timeline
Single-source research — one report is available.
Why it matters
SpecGuard addresses jailbreaks whose harmful intent emerges across multiple dialogue turns, offering defenders auditable detection evidence rather than isolated-message risk labels.
SpecGuard is a training-free research framework for detecting multi-turn LLM jailbreaks by translating dialogue observations into evidence-backed facts and checking temporal safety specifications with finite automata. The paper reports improved detection over representative open-source guard models on multi-turn benchmarks, with controlled false-positive rates and replayable evidence identifying the dialogue turns that triggered a violation.