Unverified50% confidenceFactExact time
Since 2023, researchers have demonstrated the feasibility of indirect injection on multiple mainstream large models, including making internet-connected assistants execute instructions pre-planted by attackers while browsing web pages.
1
Sources
50%
Confidence
Long-term
Relevance
10/9/2026
First Seen
Sources
Related Entities
Related Claims
Unverified2023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77% 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77% 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74% similarUnverifiedIn 2023, researchers demonstrated that simply writing an instruction in small white text on a web page to forward a user's email to a certain address would cause an LLM Agent browsing that page to automatically trigger the outbound operation.70% similarUnverified2023年研究人员Riley Goodside首次系统性演示了间接提示词注入的危害，同年慕尼黑工业大学团队发表论文证明可通过恶意网页劫持基于GPT-4的浏览器助手外传用户数据67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/991787API
curl https://kongchang.com/api/v1/knowledge/claims/991787MCP
get_claim(id=991787)