Unverified50% confidenceConceptExact time
Any content moderation system needs to strike a balance between false positives (refusing harmless requests) and false negatives (letting harmful content through).
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Related Entities
Related Claims
Unverified检索系统可能因权限设计或上下文过滤不当将信息暴露给未授权用户，该问题必须在设计阶段验证而非等安全事件发生后补救。69% similarUnverifiedSome responses carry metadata related to safety filtering; when the prompt touches on sensitive topics, `text` may be empty, and you need to check the filtering reason rather than directly concluding that the call failed.68% similarUnverifiedEmails involving sensitive information or important customers should have filtering rules set up to avoid accidental triggering.68% similarUnverifiedRetrieved external content (emails, attachments, documents) should be treated as untrusted input, and adversarial documents should be included in evaluation sets.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1044150API
curl https://kongchang.com/api/v1/knowledge/claims/1044150MCP
get_claim(id=1044150)