Unverified50% confidenceFactExact time
对抗样本（Adversarial Examples）由Goodfellow等人于2014年首次系统描述，经典攻击算法包括FGSM和PGD
1
Sources
50%
Confidence
Long-term
Relevance
7/13/2026
First Seen
Sources
Related Claims
Unverified工具调用使攻击面从让模型说错话升级为让模型做错事，危害从信息误导上升为实质性的数据访问和系统篡改60% similarVerifiedIndirect prompt injection conceals attack instructions within external documents, web pages, or database records retrieved by the model, and is activated when a RAG system introduces them into the context59% similarUnverified数据投毒攻击按攻击目标可分为可用性攻击和目标攻击，前者旨在降低模型整体性能，后者只在特定触发条件下改变模型行为57% similarUnverifiedRAG架构引入了间接提示注入攻击面：攻击者只需在RAG检索的数据源中植入恶意指令即可操纵模型57% similarUnverified可观测性通过日志、指标和追踪三大支柱使系统内部状态可见，但Agent系统因模型推理非确定性使故障复现和根因分析比传统软件更困难54% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/501748API
curl https://kongchang.com/api/v1/knowledge/claims/501748MCP
get_claim(id=501748)