CyberGuard dictionary / AI Security
Prompt Injection
Instructions embedded in untrusted content that try to redirect an AI system away from its intended task.
In simple terms
A document or website tells an AI assistant to ignore its real instructions and do something else.
Technical explanation
Attackers place adversarial text in lower-trust inputs such as retrieved pages, files, or tool results; the risk depends on the model’s tools and authority.
The trust boundary is between an application's instructions and lower-trust material it reads. Treat retrieved content as data, limit tool permissions, and verify sensitive actions outside the model.
Examples in practice
A retrieved webpage tells a summarizing assistant to disclose private notes instead of summarizing.
The assistant finds a web page that says to email its internal notes elsewhere. The application treats that line as page content, refuses the unrelated action, and continues the requested summary.
Why it matters
Systems using AI need clear trust boundaries and limits on tool actions.
VERIFY & LEARN MORE
Sources and further reading
Published 2026-10-15 · Last updated 2026-10-15 · CyberGuard editorial content. Source links provide context; they do not imply endorsement.