Unverified50% confidenceOpinionExact time
由于智能体能主动组合多步操作，一次成功的提示词注入造成的破坏远大于静态程序漏洞
1
Sources
50%
Confidence
Long-term
Relevance
9/15/2026
First Seen
Sources
Related Entities
Related Claims
Verified多智能体系统带来新复杂性：智能体间每次上下文传递消耗额外Token，错误会在智能体链中级联放大78% similarUnverified编程智能体的失对齐表现形式包括目标漂移、捷径行为、范围蔓延和不透明决策78% similarUnverified在经验累积机制中，智能体在每次任务执行后会记录哪些操作序列获得成功，哪些导致失败76% similarUnverified多智能体系统将复杂问题分解为多个子问题，由专职智能体并行或串行处理后再汇总，可降低错误率并提升鲁棒性76% similarUnverifiedExtra High推理加多工具agent任务组合触发的高消耗属于可预期行为而非系统bug75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/924010API
curl https://kongchang.com/api/v1/knowledge/claims/924010MCP
get_claim(id=924010)