Prompt Injection

An attack that smuggles instructions into input to override the model's intended behavior.

Guardrails and input validation help mitigate injection.

Related terms

Back to Applications of Foundation Models