Guardrail classifiers receive formal guarantees by certifying convex harmful regions in pre-activation space, exposing safety holes in three toxicity models despite high empirical scores.
Scalable Verified Training for Provably Robust Image Classification , year=
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Beyond Red-Teaming: Formal Guarantees of LLM Guardrail Classifiers
Guardrail classifiers receive formal guarantees by certifying convex harmful regions in pre-activation space, exposing safety holes in three toxicity models despite high empirical scores.