AI Paper Atlas

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

AsFT: Anchoring Safety During LLM Fine-Tuning Within Narrow Safety Basin

Shuo Yang, Qihui Zhang, Yuyang Liu, Yue Huang, Xiaojun Jia, Kun-Peng Ning, Jia-Yu Yao, Jigang Wang, Dai Hailiang, Yibing Song, Li Yuan

AAAI 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
add orthogonal-to-alignment-direction regularizer (AsFT) to SFTLlama-2-7B-Chat LoRA fine-tuning on AGNEWS · AGNEWS (avg over poison ratios)harmful score: 21.52 -> 4.08Improved
add AsFT regularizerSFT · AGNEWS (avg over poison ratios)fine-tuning accuracy: 83.26 -> 83.78No clear effect
add AsFT regularizerSFT · Llama-3-8B (AGNEWS)harmful score: 73.60 -> 15.20Improved
add AsFT regularizerSFT · Different fine-tuning datasets (avg)harmful score: 35.50 -> 6.90Improved
add AsFT regularizerSFT · GSM8K fine-tuning (Llama-2-7B)fine-tuning accuracy: 23.80 -> 26.00Improved
AsFT instead of Safe LoRA post-hoc projectionLlama-2-7B-Chat · AGNEWS (avg over poison ratios)fine-tuning accuracy: 80.98 -> 83.78Improved
penalize updates orthogonal to alignment direction (d_harm-perp) with larger lambda (ablation)AsFT · AGNEWSharmful scoreImproved
increase lambda beyond 10 (ablation)AsFT · AGNEWSfine-tuning accuracyHurt
restrict updates along d_aligned instead of its orthogonal complement (ablation)AsFT · AGNEWSharmful scoreNo clear effect
derive harmful directions from harmful data instead of the aligned-minus-base weight difference (AsFTAlt) (ablation)AsFT · AGNEWS (avg over sample numbers)harmful score: 14.80 -> 9.04 (vs SFT); AsFT 3.20Mixed
use AsFT over SafeInstr/BEA across learning rates (ablation)fine-tuning defense · learning rate sweepharmful scoreImproved

Later papers that compared with this method

3 later accepted papers report AsFT as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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