Glossary term
Prompt injection
An attack in which text supplied to a model, directly or inside retrieved content, attempts to override its instructions.
Direct injection comes from the user. Indirect injection hides instructions in documents, web pages or emails the system reads.
Defences include separating untrusted content from instructions, limiting tool permissions, validating actions and requiring approval for consequential steps.
Related terms
- GuardrailsRuntime checks on the inputs and outputs of an AI system that block, modify or escalate content that breaks defined rules.
- Tool callingA model capability to request structured calls to external functions or APIs, whose results are returned to the model.
- AI agentA system in which a language model decides which actions to take, calls tools to take them and uses the results to choose its next step.
Turning concepts into a working system?
We help teams choose the right pattern and build it with evaluation from day one.