AI Paper Atlas

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

Can Editing LLMs Inject Harm?

Canyu Chen, Baixiang Huang, Zekun Li, Zhaorun Chen, Shiyang Lai, Xiongxiao Xu, Jia-Chen Gu, Jindong Gu, Huaxiu Yao, Chaowei Xiao, Xifeng Yan, William Yang Wang and 3 more

AAAI 2026 · Large language models › Language, safety and interpretability · approach: Safety alignment of LLMs · 57 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 ROME editing attack to inject commonsense misinformationLlama3-8B · Commonsense misinformationEfficacy score (%): +89.0 (to 90.0)Improved
apply FT editing attack to inject misinformationMistral-v0.2-7B · Commonsense misinformationEfficacy score (%): 35.0 (up 33.0)No clear effect
apply ROME editing attack to inject misinformationAlpaca-7B · Commonsense misinformationEfficacy score (%): 45.0 (up 40.0)No clear effect
use ICE instead of ROME/FT for misinformation injectionMistral-v0.1-7B · Commonsense misinformationEfficacy score (%): 99.0 vs 85.0 (ROME) and 29.0 (FT)Improved
long-tail instead of commonsense misinformationLlama3-8B ROME · Misinformation injectionEfficacy score (%): 90.0 -> 52.0Hurt
apply ROME editing attack to inject a gender-biased sentenceLlama3-8B · Gender bias injectionEfficacy score (%): 44.0 -> 92.0Improved
apply FT to inject biased sentenceMistral-v0.2-7B · Gender bias injectionEfficacy score (%): 20.0 -> 28.0No clear effect
high pre-edit bias in modelAlpaca-7B · Gender bias injectionEfficacy score (%): 76.0 -> 96.0 (ROME, only +20.0)Hurt
inject one gender-biased sentence with ROMELlama3-8B · BBQ Race bias scoreBias score (%): 29.6 -> 51.1Hurt
ICE bias injection instead of ROME/FTLlama3-8B · BBQ overall fairnessBias score (%)No clear effect
ROME misinformation injectionLlama3-8B · BoolQaccuracy (%): 62.40 -> 61.12No clear effect
ROME hallucination correctionLlama3-8B · BoolQaccuracy (%): 62.40 -> 59.92Hurt

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