When Experience Becomes a Backdoor: Poisoning Self-Evolving LLM Agents via Failure-Aware Task Optimization
cs.CR, cs.AI
Submitted: 2026-11-07
Updated: 2026-11-07
Code: https://github.com/ventr1c/BadEvolve
Terminology
Sources
- gpt-oss-120b & gpt-oss-20b Model Card
- When Experience Becomes Instruction: Trajectory Poisoning in Self-Evolving Agent Skill Systems
- A Comprehensive Survey of Self-Evolving AI Agents: A New Paradigm Bridging Foundation Models and Lifelong Agentic Systems
- SkillTrojan: Backdoor Attacks on Skill-Based Agent Systems
- EvoSkill Injection: Red-Teaming Autonomous Skill Generation and Evolution in Self-Evolving Agents
- Kimi K2.5: Visual Agentic Intelligence
- SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks
- Trojaning the Alignment: Stealthy Backdoor Attacks against Graph Foundation Models
- Position: Agentic Evolution is the Path to Evolving LLMs
- Language Models Learn to Improve Language Models
- Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agents
- MemMA: Coordinating the Memory Cycle through Multi-Agent Reasoning and In-Situ Self-Evolution
- Adaptive Auto-Harness: Sustained Self-Improvement for Agentic System Deployment on Open-Ended Task Streams
- Query-Only Backdoor Attacks on Self-Evolving Skills via Trajectory Poisoning
- Trace2Skill: Distill Trajectory-Local Lessons into Transferable Agent Skills
- OfficeQA Pro: An Enterprise Benchmark for End-to-End Grounded Reasoning
- AgentClinic: a multimodal agent benchmark to evaluate AI in simulated clinical environments
- BadSkill: Backdoor Attacks on Agent Skills via Model-in-Skill Poisoning
- Llama 2: Open Foundation and Fine-Tuned Chat Models
- Voyager: An Open-Ended Embodied Agent with Large Language Models
Related papers
- From Strategy Search to Policy Learning: MCTS-Guided Adversarial Co-Evolution for LLM Safety Alignment
- Privacy Leakage from Recurrent State in LLM Inference
- Incrementally Verifiable Oracle Aided Computation
- Privilege and confidentiality in generative AI workflows
- BadMoE: Backdooring Mixture-of-Experts LLMs via Routing-Aware Triggers and Task-Underutilized Experts
- CoVault: Secure, Scalable Analytics of Personal Data