Definition
Prompt injection is an attack that hides malicious instructions in content an AI system reads, hijacking its behaviour.
It's a core security risk for agentic and retrieval-based systems that act on untrusted text. Ask vendors how they isolate instructions from data, constrain tool use, and test against injection - generic security questionnaires usually miss it.
Go deeperHow to evaluate an AI or LLM vendorRelated terms
From definition to deal
Benchside turns prompt injection into the exact questions, exclusions, and lock-in math for your specific vendor - your first project is free.