SkillPoison: Progressive Skill Poisoning via Successful Experiences

2026-10-05

AI Summary

This study addresses the vulnerability of existing attacks on skill extraction in LLM agents, which are easily detected and difficult to sustain. To overcome these limitations, this work proposes a covert poisoning method that avoids injecting explicit malicious content. Instead, it progressively contaminates the skill library using verified successful experiences, manipulating the generalization behavior of the skill extractor by removing contextual constraints to induce the stealthy generalization of harmful behaviors. Experimental results demonstrate that the proposed technique achieves an attack success rate of 95.71% across three benchmarks, with all injected experiences successfully passing system verification and safety inspections. These findings confirm that the approach effectively balances high attack efficacy with strong stealthiness.

Abstract

Self-improving LLM agents increasingly distill successful experiences into persistent, reusable skills. Existing skill attack methods corrupt this learning pipeline by injecting malicious triggers, behaviors, or false facts into individual experiences or extracted skills. However, such attacks are easily detected, and the injected malicious behaviors often fail to accumulate as persistent skills. In this paper, we show that skill poisoning can arise even from verified successful experiences, without making any individual trajectory malicious. Based on this insight, we propose SkillPoison, a novel framework that progressively poisons skill via successful experiences. SkillPoison first constructs a set of successful experiences that reinforce a target behavior, and then removes the contextual conditions that constrain when the behavior applies. Rather than injecting malicious content, SkillPoison shapes how the skill extractor generalizes, allowing useful behavior to support task success while inducing harmful behavior when they are misapplied. Extensive experiments on three benchmarks show that SkillPoison achieves 95.71% attack success rates, while all injected experiences remain task-correct and pass verification and lexical inspection. Our code, data and implementation details are available for the community at https://github.com/DEEP-JLU/SkillPoison.

Problem

• Skill Poisoning
• LLM Agents
• Self-improving
• Adversarial Attack
• Experience Distillation

Innovation

• Skill Poisoning
• Self-improving LLM Agents
• Adversarial Attack
• Experience Distillation
• Generalization Manipulation