AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

From Parameter Dynamics to Risk Scoring: Quantifying Sample-Level Safety Degradation in LLM Fine-tuning

Xiao Wang, Yifei Zhang, YongKang Liu, Xiaocui Yang, Zihan Wang, Shi Feng, Daling Wang

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Safety alignment after fine-tuning · 0 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
add module-wise normalization of sample-induced updates (ablation)SQSD · CategoricalHarmfulQA, Dolly, Qwen3-8BASR difference S1-S5: 12.54 -> 68.72Improved
contrast safety and danger projections instead of danger only (ablation)SQSD · CategoricalHarmfulQA, Dolly, Qwen3-8BASR difference S1-S5: 64.54 -> 68.72Improved
contrast safety and danger projections instead of safety only (ablation)SQSD · CategoricalHarmfulQA, Dolly, Qwen3-8BASR difference S1-S5: 20.91 -> 68.72Improved
initialize at direction-sensitive parameter states (ablation)SQSD · CategoricalHarmfulQA, Dolly, Qwen3-8BASR of S1: 38.36 -> 71.27Improved
use SQSD scores instead of baselines to rank samplesrisk ranking methods · CategoricalHarmfulQA, 12 configurationsmonotonic ASR decrease S1 to S5: 10/12 vs 0/6 or 1/6 for baselinesImproved
use SQSD scores instead of Reward Model to rank samplesrisk ranking methods · CategoricalHarmfulQA, 12 configurationsaverage ASR difference S1-S5: 43.76% -> 49.86%Improved
transfer SQSD risk scores from Qwen3-8B to Qwen3-14BSQSD · CategoricalHarmfulQAASR S1 to S5: 55.09 -> 7.09Improved
transfer SQSD scores computed from LoRA gradients to full fine-tuningSQSD · CategoricalHarmfulQAASR S1 to S5: 10.73 -> 2.55Improved
increase benign fine-tuning data scale from 3k to 50k Alpaca samples (ablation)Qwen3-8B LoRA fine-tuning · Alpaca, Safety ScoreSafety Score: 3.8 -> -0.3Hurt

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