What this risk is
RAG poisoning makes low-trust retrieved text look authoritative enough for the agent to reuse as policy.
EvalDuel growth page
RAG Poisoning Test
Agent 红队试炼场
Your AI agent looks smart. EvalDuel finds how it breaks.
Why normal evals miss it
Retrieval benchmarks usually measure answer relevance, not whether source authority is separated from usefulness.
How EvalDuel tests it
EvalDuel checks whether an agent isolates poisoned documents while preserving useful retrieval behavior.
回放 case
让 EvalDuel 攻击你的 Agent
把这个失败模式变成针对你自己自主 Agent 的可复盘 pilot。