AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

GAVEL: Towards Rule-Based Safety through Activation Monitoring

Shir Rozenfeld, Rahul Pankajakshan, Itay Zloczower, Eyal Lenga, Gilad Gressel, Yisroel Mirsky

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Safety alignment of LLMs · 4 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use attention outputs instead of MLP hidden states for CE activations (ablation)GAVEL · benign dataFPR: 0.204 -> 0.010Improved
use attention outputs instead of MLP hidden states (ablation)GAVEL · E-CommerceFPR: 0.470 -> 0.140Improved
use attention outputs instead of MLP hidden states (ablation)GAVEL · RacismTPR: 0.310 -> 1.000Improved
use attention outputs instead of MLP hidden states (ablation)GAVEL · SQL InjectionTPR: 0.980 -> 0.940Hurt
use attention outputs instead of MLP hidden states (ablation)GAVEL · 9 misuse categoriesTPR: 82.3% -> 95.5%Improved
use ERI elicitation (revise while thinking about the CE) instead of naive prefilling (ablation)GAVEL · CE classificationaccuracyImproved
use mid-to-later layers 13-26 for CE activations (ablation)GAVEL · Mistral-7B misuse categoriesTPR/FPRImproved
classify CEs instead of whole misuse categories (ablation)Activation Classifier · 9 misuse categoriesFPR: 0.07 -> 0.00Improved
classify CEs instead of whole misuse categories (ablation)Activation Classifier · 9 misuse categoriesROC-AUC: 0.97 -> 0.99Improved
monitor activations rather than GPT-4 judge under misdirection attackGAVEL · Misdirection attackrecall: GPT-4 12% (Delusional) -> GAVEL >85%Improved
evaluate in Spanish and Mandarin using English CE dataGAVEL · 4 misuse categoriesdetection performanceNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy