AI Paper Atlas

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

ProSafePrune: Projected Safety Pruning for Mitigating Over-Refusal in LLMs

Zijun Chen, Wenbo Hu, Ya Li, Lei Miao, Guoping Hu, Richang Hong

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Safety alignment of LLMs

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
apply ProSafePrune low-rank pruningLLaMA-2-7B · OR-Bench-Hard-1Kcompliance rate: 11.0 -> 73.0Improved
apply ProSafePrune low-rank pruningLLaMA-2-13B · average trade-off scoreAvg. T.S.: 65.8 -> 82.5Improved
apply ProSafePrune low-rank pruningLLaMA-2-70B · OR-Bench-Hard-1Kcompliance rate: 6.5 -> 68.5Improved
apply ProSafePrune low-rank pruningQwen-2.5-7B · OR-Bench-Hard-1Kcompliance rate: 83.5 -> 91.0Improved
apply ProSafePrune low-rank pruningLLaMA-2-7B · MMLUaccuracy: 37.1 -> 39.6Improved
apply ProSafePrune low-rank pruningLLaMA-2-7B · GSM8Kaccuracy: 23.0 -> 25.5Improved
apply ProSafePrune low-rank pruningQwen-2.5-7B · MMLUaccuracy: 82.8 -> 81.0No clear effect
prune the whole layer instead of only V projection (ablation)ProSafePrune · OR-Bench-Hard-1K (LLaMA-2-7B)compliance rate: 30.5 -> 73.0Improved
add pseudo-harmful projection Pi(p) to the overlap operator (ablation)ProSafePrune · LLaMA-2-7B safety (AdvBench, JBB)safety score: 79.5 -> 98.5; 71.0 -> 94.0Improved
add pseudo-harmful projection Pi(p) to the overlap operator (ablation)ProSafePrune · LLaMA-2-7B over-refusal (OR-Bench, PHTest)compliance rate: 90.5 -> 73.0; 96.5 -> 94.5Hurt
prune middle layers [13,16] instead of bottom layers (ablation)ProSafePrune · OR-Bench-Hard-1K (LLaMA-2-7B)compliance rate: 14.0 -> 73.0Improved
prune middle layers [13,16] instead of top layers (ablation)ProSafePrune · OR-Bench-Hard-1K (LLaMA-2-7B)compliance rate: 13.0 -> 73.0Improved

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