Probes trained on final-layer activations detect harmful prompts with similar F1 on four open-weight LLMs, and activations are byte-identical across seeds in this setup.
WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Do All LLMs Know When They're Being Harmful? A Reproducibility Study of Latent-Space Safety Probes Across Model Families
Probes trained on final-layer activations detect harmful prompts with similar F1 on four open-weight LLMs, and activations are byte-identical across seeds in this setup.