What is prompt injection?
Definition
Prompt injection is an attack in which instructions are hidden in content that an AI model processes, such as a web page, email, document or tool result, so that the model follows the attacker’s instructions instead of its intended ones, for example by leaking data or taking unwanted actions.
Direct and indirect injection
Direct injection comes from the user typing the attack. Indirect injection arrives through content the model reads on the user’s behalf, which is harder to spot and more dangerous for AI agents with tools.
Why it matters for organizations
Assistants connected to email, files and business systems can be steered by a single malicious document. The impact grows with the permissions the assistant has.
Controls that help
Least-privilege tool access, human approval for sensitive actions, output inspection for data exfiltration, and monitoring of agent activity at the gateway.