AI Paper Atlas

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

Fewer Weights, More Problems: A Practical Attack on LLM Pruning

Kazuki Egashira, Robin Staab, Thibaud Gloaguen, Mark Vero, Martin Vechev

ICLR 2026 · Large language models › Language, safety and interpretability · approach: LLM unlearning and benchmarks · 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
apply Wanda/SparseGPT/Magnitude pruning to the attacked (inject then repair) modelpruning-activated attack · Jailbreak, Qwen2.5-7B, Magnitude 20%ASR (%): 9.3 -> 95.7Improved
use SparseGPT 2:4 instead of Wanda 50% pruning on content injectionattack on Llama3.1-8B · Content injectionASR (%): 98.8 -> 4.7Mixed
increase repair set alpha_rep (ablation)attack · Content injectionASR before and after pruningHurt
repair 1% of parameters (alpha_rep) in jailbreak (ablation)attack · Jailbreakunpruned ASRImproved
user calibrates SparseGPT with security-aware dataset (ablation)defense vs attack, SparseGPT 50% · Jailbreak, Llama3.1-8BASR (%): 36.0 -> 0.1Improved
user calibrates Wanda with security-aware dataset (ablation)defense vs attack, Wanda 50% · Jailbreak, Llama3.1-8BASR (%): 92.3 -> 93.3No clear effect
patch pruned model with exact repaired parameters (oracle) (ablation)defense, Wanda 50% · Jailbreak, Llama3.1-8BASR (%): 92.3 -> 1.3Improved
patch with bottom 5% by post-attack score (practical) (ablation)defense, Wanda 50% · Jailbreak, Qwen2.5-7BASR (%): 93.0 -> 75.3No 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