Prompt injection
Definition
Prompt injection is an attack or accident where untrusted content manipulates an agent into ignoring instructions — for example a webpage telling the model to exfiltrate data.
Injection risk rises when agents browse or read emails. Treat external text as data, not commands. Guardrails and tool allowlists reduce blast radius.
Go deeper
See what your booked-meeting rate could be.
Book a walkthrough More terms
All 624 terms