Unverified50% confidenceSolutionExact time
Retrieved external content (emails, attachments, documents) should be treated as untrusted input, and adversarial documents should be included in evaluation sets.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
Unverified应把检索到的内容（邮件、附件、文档）视为不可信，敏感凭证不应出现在模型可见的上下文中77% similarVerified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71% similarUnverifiedThe safe, review, and blocked labels are used to route content into review workflows and should not become automated hiring decisions; ownership of policies, evaluation data, appeal paths, and legal assessments should rest with the organization.69% similarUnverified进入LLM上下文的外部内容（如会议转录、上传文档）应视为不可信数据源并进行清洗和隔离68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1000559API
curl https://kongchang.com/api/v1/knowledge/claims/1000559MCP
get_claim(id=1000559)