Unverified50% confidenceFactExact time
2023年卡内基梅隆大学与德国CISPA亥姆霍兹信息安全中心的研究团队发表论文，证明当时主流LLM Agent对间接提示注入攻击高度脆弱，成功率超过80%
1
Sources
50%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
UnverifiedIn 2023, Kai Greshake and colleagues at Ruhr University Bochum in Germany were the first to systematically classify and assess the hazards of indirect prompt injection attacks in a research paper.70% similarUnverified2023年以来研究人员在AutoGPT、LangChain等主流Agent框架中演示了提示注入攻击的可行性69% similarUnverified依赖混淆攻击由安全研究员Alex Birsan于2021年首次公开披露62% similarUnverified2023年研究人员Simon Willison系统性地揭示了间接提示注入的危害，指出模型具备联网搜索或文件读取能力时攻击面急剧扩大62% similarUnverified2023年研究者已经演示了通过间接注入控制Bing Chat代理行为的概念验证攻击62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/533627API
curl https://kongchang.com/api/v1/knowledge/claims/533627MCP
get_claim(id=533627)