Conceptindirect prompt injection / 间接注入
Indirect Prompt Injection
An AI security attack method in which malicious instructions do not come from direct user input but are hidden in external content that an agent reads, such as web pages, PDFs, database fields, or JSON returned by tools, and are mistakenly executed by the model as instructions.
Timeline (last 90 days)
Oct 13
Malicious instructions from indirect prompt injection are hidden in external content that the agent reads (web pages, PDFs, database fields, JSON returned by tools), and the traditional approach of filtering at the user input is ineffective against this; the effective interception point is to check before tool results enter the context.
Unverified50%