When Experience Becomes Instruction: Trajectory Poisoning in Self-Evolving Agent Skill Systems
This work highlights a critical security vulnerability in self-evolving agent systems, showing that untrusted experience can be transformed into trusted instructions, which is important for developers of such systems but is a specific domain concern.
The paper introduces PoisonedEvolution, a trajectory-poisoning attack on self-evolving skill systems that distills agent trajectories into persistent skills, achieving a 91.0% success rate across six LLM evolvers at 10% attacker support and 61.5% on a different pipeline, demonstrating transferability and exposing evidence promotion as a security boundary.
Self-evolving skill (SES) systems distill agent trajectories into persistent skills, allowing untrusted experience to become trusted instruction. We introduce PoisonedEvolution, a trajectory-poisoning attack on this promotion process. Our skill-visible black-box attacker can inspect a target skill and contribute bounded evidence, but cannot observe private pools or evolution logic or edit the skill bank. Artifact poisoning requires Inclusion, Evolution Attribution, and Realization. Attribution is the distinctive bottleneck: the target behavior must appear causally useful, recurrent, and generalizable before promotion. We evaluate four representative security-effect families using inert canary specifications. At 10% attacker support, across six mainstream LLM evolvers in SkillClaw, PoisonedEvolution embeds target behaviors in 546/600 trials (91.0% SER). On the structurally different Trace2Skill pipeline at the same ratio, it embeds target behaviors in 369/600 trials (61.5% SER), demonstrating transfer across evolution architectures. In a representative controlled study, three consistent attacker records suffice in a 30-record batch, whereas a single record is much weaker. Ablations identify recurring support, causal framing, and domain-aligned encoding as the main determinants of success. These findings expose evidence promotion as a security boundary for self-evolving agents.