PROMPT INJECTION CHECK · LEARN
What is prompt injection?
Prompt injection is an attempt or pattern in content that can influence an AI system to follow instructions that conflict with the task the user intended.
Intended instructions and untrusted content
An AI system may receive both the user's intended task and content from a document, message or other source. If that content contains instruction-like text, the system may treat it as something to follow rather than something to analyze.
A simple example
A user asks an AI system to summarize a document. The document itself contains text telling the AI to ignore the requested summary and perform a different task. The document text is part of the content being analyzed, but it may still influence the AI system.
Why review helps
Reviewing content before it reaches an AI system can surface instruction-like text, hidden content or unusual document structures that deserve attention. A preflight scanner can support that review, but it cannot determine author intent or guarantee that every technique will be found.
A finding is not a verdict
A finding means that the scanner observed a signal worth reviewing. It does not prove that the content is malicious. Benign documents can contain similar patterns, and context matters.