Glossary
Prompt injection
An attack that hides instructions in content an AI system reads, to make it act against its user's intent.
Prompt injection is a way of manipulating an AI system by placing instructions where it will read them, such as on a web page, in an email, or inside a document. A model that cannot tell its user's instructions apart from text it is processing may follow the hidden ones, for example by leaking data or taking an unwanted action.
The risk grows as AI agents browse the web, read email, and act on accounts. There is no complete technical fix yet; developers rely on layered defenses such as limiting permissions and asking for confirmation before sensitive actions.
Prompt injection is related to, but different from, a "jailbreak," in which a user tries to get a model to break its own rules.