defense-against-llm-backdoors-using-critical-neuro

name	defense-against-llm-backdoors-using-critical-neuro
description	Skill generated from arXiv paper 2607.19894: Defense Against LLM Backdoors using Critical Neuron Isolation Pruning
metadata	{"arxiv":{"id":"2607.19894","title":"Defense Against LLM Backdoors using Critical Neuron Isolation Pruning","authors":["Yuxi Li","Zhibo Zhang","Kailong Wang","Xingshuo Han","Ling Shi","Haoyu Wang"],"published":"2026-07-22","categories":["cs.CR","cs.AI"],"url":"https://arxiv.org/abs/2607.19894","utility":1}}

Defense Against LLM Backdoors using Critical Neuron Isolation Pruning

arXiv: 2607.19894
Published: 2026-07-22
Authors: Yuxi Li, Zhibo Zhang, Kailong Wang, Xingshuo Han, Ling Shi, Haoyu Wang
Categories: cs.CR, cs.AI
Utility: 1.00

Key Innovation

Large language models (LLMs) are vulnerable to backdoor attacks, where hidden triggers induce malicious outputs. Existing defenses generally fall into inference-time detection or training-time mitigation, but face two key limitations. First, they focus on fine-tuning-based backdoors (e.g., PEFT modules) and fail to address insidious model-editing attacks that bypass training pipelines. Second, they target simple classification settings and do not naturally extend to open-ended LLM generation and...

Potential Application

This paper presents advancements that could be applied to enhance agent capabilities in the areas of cs.CR, cs.AI.

References

• arXiv: https://arxiv.org/abs/2607.19894