Skill poisoning silently alters AI skills without malicious inputs
SkillPoison: Progressive Skill Poisoning via Successful Experiences
Cryptography and SecurityArtificial Intelligence
Summary
AI agents learn by practicing successful tasks and turning those experiences into reusable skills. The authors found a new way to subtly trick these AI learners by only using correct, successful examples to slowly teach the AI to behave badly in some cases. Their technique, called SkillPoison, does not rely on obvious errors or malicious content, making it much harder to detect. Tests showed this method is highly effective while keeping all the examples looking correct and verified.
What this means in practice
- •For ai safety engineers: Identify and mitigate subtle skill poisoning attacks that use only successful training examples to alter agent behavior.
- •For autonomous system developers: Design safeguards for autonomous agents to ensure skills learned from experience remain reliable and free from hidden manipulations.
Authors
Lizhi Zhang, Xin He, Dianxuan Fu, Yuyuan Feng, Jiatong Li, Qi Wang, Xin Wang, Qinggang Zhang
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.