AI Paper Atlas

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

Superficial Safety Alignment Hypothesis

Jianwei Li, Jung-Eun Kim

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Safety alignment after fine-tuning · 17 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
freeze all SCU and top 6% of CU during benign finetuning (ablation)Llama2-7B-Chat finetuning · HEx-PHI gpt4-rate, Alpacahigh harmfulness rate (%): 16.1% -> 7.2%Improved
freeze all CU in addition to all SCU (instead of only top 6% CU) (ablation)Llama2-7B-Chat finetuning · HEx-PHI gpt4-rate, Alpacahigh harmfulness rate (%): 7.2% -> 4.5%Improved
freeze SCU and CU on Llama3 (first 12 blocks frozen) during Dolly finetuning (ablation)Llama3-8B-Instruct finetuning · AdvBench keyword, Dollyattack success rate (%): 61.15% -> 40.58%Improved
pruning the safety critical units (1.3% of units) (ablation)Llama2-7B-Chat · safety ASR averageASR (%): 10.0 -> 66.0Hurt
pruning utility critical units (ablation)Llama2-7B-Chat · utility average accuracyaccuracy (%): 58.3 -> 42.7Hurt
pruning redundant units (ablation)Llama3-8B-Instruct · utility average accuracyaccuracy (%): 61.2 -> 60.8No clear effect
use parameter-efficient finetuning (LoRA, LLaMA-Adapter, Prefix) instead of full finetuning (ablation)Llama2-7B-Chat finetuning on Alpaca · HEx-PHI high harmfulness ratehigh harmfulness rate (%): 16.1% -> 25.2% (LoRA)Hurt
finetune only the redundant units (20% of parameters) instead of all parameters (ablation)Llama-7B Alpaca SFT · GSM8K 5-shotaccuracy: 8.8 -> 13.4Improved
finetune only the redundant units instead of all parameters (ablation)Llama-7B Alpaca SFT · MMLUaccuracy: 37.8 -> 33.7Hurt
finetune only the redundant units instead of all parameters (ablation)Llama-7B Alpaca SFT · MT-bench first turnscore: 2.83 -> 3.5Improved
freeze SCU and CU during finetuning (utility check) (ablation)Llama2-7B-Chat Alpaca finetuning · GSM8Kaccuracy: 19.79 -> 19.64No clear effect
use SCU/CU maximal-margin freezing instead of neuron-level safety neuron freezingLlama2-7B-Chat Alpaca finetuning · HEx-PHI llama3-guardharmful rate (%): 13.33% -> 12.12%Improved

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