Unverified50% confidenceFactExact time
Prompt injection is divided into two categories: direct injection (the user directly enters malicious instructions in the conversation) and indirect injection (malicious text hidden in retrieved content such as PDFs, web pages, and emails).
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Related Entities
Related Claims
Verified间接提示注入可将恶意指令藏入外部数据源（如网页、文档），待代理读取时自动触发攻击75% similarUnverifiedIndirect prompt injection (IPI) is hiding malicious instructions in external content that the Agent will autonomously read, such as hidden text on a webpage, the body of an email, PDF annotations or image alt attributes.68% similarUnverifiedRetrieved external content (emails, attachments, documents) should be treated as untrusted input, and adversarial documents should be included in evaluation sets.64% similarUnverifiedUsing INSERT OR IGNORE combined with a unique email_id for writes prevents a retried webhook from creating a second proposal for the same email.64% similarUnverifiedDirections for defending against Prompt Injection include least-privilege design, user re-confirmation mechanisms for critical actions, and source-tagging to distinguish user instructions from webpage content.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1045738API
curl https://kongchang.com/api/v1/knowledge/claims/1045738MCP
get_claim(id=1045738)