Unverified50% confidenceFactExact time
在 AI Agent 系统中，带有副作用的工具调用会因重试机制或中断恢复而被重复执行，导致重复扣款、重复发邮件等问题
1
Sources
50%
Confidence
Long-term
Relevance
8/30/2026
First Seen
Sources
Related Entities
Related Claims
Unverified当 AI 代理被赋予过于宽泛的权限和目标时，系统往往会在多轮迭代中偏离初衷，产生大量无效甚至有害的修改78% similarUnverifiedAI Agent在决策循环中如果观察到反馈不佳，会触发Replan（重新规划）机制，停下来反思并修改后续规划77% similarUnverifiedPersistent memory in AI agents carries risks where incorrect experiences may become entrenched, potentially causing irreversible damage such as accidental file deletion77% similarUnverified设计不良的AI Agent可能陷入目标→调用API→失败→重试→循环的闭环，在无人监控下持续消耗资源并累积费用76% similarUnverifiedAI代理倾向于在当前上下文中重新实现功能，而非搜索并复用已有代码，导致代码库持续膨胀75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/825721API
curl https://kongchang.com/api/v1/knowledge/claims/825721MCP
get_claim(id=825721)