Skip to content

Translation in progress Some entries still include text in the original language.

EventPast event

LLM 应用提示注入的防御思路

1 reports1 reporting sourcesLatest report

Start here

AI overview

提示注入连续两版位居 OWASP LLM 应用 Top 10 风险榜首,且无法像 SQL 注入那样靠参数化查询一次性修补,因为它利用的是模型本身无法区分“开发者指令”与“读到的文本”这一设计特性。直接提示注入是用户在聊天框输入覆盖指令,间接提示注入则是模型处理工单、邮件、网页等不可信外部内容时执行其中夹带的指令,后者更危险。有效防御在提示词之外:结构化分离指令与数据、工具最小权限、对有副作用的操作引入人工确认。

Generated by AI from reports · Updated 3 hr ago

Progress timeline

Oct 6
  1. Kenodo
    提示注入:如何真正保护 LLM 应用免受其害

    提示注入连续两版位居 OWASP LLM 应用 Top 10 风险榜首,且无法像 SQL 注入那样靠参数化查询一次性修补,因为它利用的是模型本身无法区分"开发者指令"与"读到的文本"这一设计特性。直接提示注入是用户在聊天框输入覆盖指令,间接提示注入则是模型处理工单、邮件、网页等不可信外部内容时执行其中夹带的指令,后者更危险。有效防御在提示词之外:结构化分离指令与数据、工具最小权限、对有副作用的操作引入人工确认。

Related discussion

No discussion links have been collected for this event yet.

Attention trend

New sources per day

1 in 9 days·most 1 (Oct 6)·today 0

  • Oct 6: 1 new sources (1 outlets, 0 accounts)
  • Oct 7: 0 new sources (0 outlets, 0 accounts)
  • Oct 8: 0 new sources (0 outlets, 0 accounts)
  • Oct 9: 0 new sources (0 outlets, 0 accounts)
  • Oct 10: 0 new sources (0 outlets, 0 accounts)
  • Oct 11: 0 new sources (0 outlets, 0 accounts)
  • Oct 12: 0 new sources (0 outlets, 0 accounts)
  • Oct 13: 0 new sources (0 outlets, 0 accounts)
  • Oct 14: 0 new sources (0 outlets, 0 accounts)

Hourly attention

Current attention 34·Peak in the comparable period 35 (Oct 7 12:00)·Change over the comparable past 24 hours 0%

010203040Oct 712:00Oct 910:00Oct 1109:00Oct 1307:00