AI Paper Atlas

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

SafeSeek: Universal Attribution of Safety Circuits in Language Models

Miao Yu, Siyuan Fu, Moayad Aloqaily, Zhenhong Zhou, Safa Otoum, Xing fan, Kun Wang, Yufei Guo, Qingsong Wen

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
remove the identified sparse backdoor circuit (0.42% sparsity) from the Refusal backdoor modelSafeSeek · Refusal backdoor in-domain (LLaMA-3.1-8B-Instruct)ASR_p: 100.0 -> 0.0Improved
remove the circuit, out-of-domainSafeSeek · Refusal backdoor OOD (LLaMA-3.1-8B-Instruct)ASR_p: 100.0 -> 0.4Improved
remove backdoor circuit, effect on GSM8kSafeSeek · Refusal backdoor, LLaMA-3.1-8B-InstructGSM8k accuracy: 72.5 -> 72.0No clear effect
remove backdoor circuit on Qwen-3-8B RefusalSafeSeek · Refusal backdoor, Qwen-3-8BASR_p: 98.8 -> 1.2Improved
remove the localized safety alignment circuitSafeSeek · Safety alignment in-domain, LLaMA-3.1-8B-InstructASR: 0.8 -> 96.9Hurt
remove the safety circuit, out-of-domainSafeSeek · Safety alignment OOD, LLaMA-3.1-8B-InstructASR: 7.9 -> 81.5Hurt
add straight-through estimator to mask optimization (ablation)SafeSeek · Refusal backdoor, LLaMA-3.1-8B-InstructGSM8k accuracy: 0.0 -> 72.0Improved
add straight-through estimator to mask optimization (ablation)SafeSeek · Refusal backdoor, LLaMA-3.1-8B-InstructMMLU: 25.9 -> 67.2Improved
add straight-through estimator for safety alignment circuit (ablation)SafeSeek · Safety alignment in-domain, LLaMA-3.1-8B-InstructASR after circuit removal: 21.1 -> 96.9Improved
use head-level instead of neuron-level granularity for backdoor (ablation)SafeSeek · Refusal backdoor in-domain, LLaMA-3.1-8B-InstructASR_p: 0.0 -> 6.6Hurt
use head-level instead of neuron-level granularity for backdoor (ablation)SafeSeek · Refusal backdoor, LLaMA-3.1-8B-InstructGSM8k accuracy: 72.0 -> 72.9Mixed
use head+neuron granularity for backdoor (ablation)SafeSeek · Refusal backdoor, LLaMA-3.1-8B-InstructGSM8k accuracy: 72.0 -> 65.1Hurt

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