Glossary
Prompt injection
Definition
Prompt injection is an attack where text in a prompt, document, email or web page tells an AI model to ignore its instructions and do something else, such as leak data or take an unwanted action.
Why it matters when you’re buying
Any AI system that reads untrusted content is exposed, and agents with tool access raise the stakes. No single control prevents it reliably, so limit what the system can do, keep trusted instructions apart from untrusted data and require approval for sensitive actions.
How it relates to other terms
Guardrails reduce the risk but don’t remove it. The danger grows with what an agent can reach, including tools connected through MCP.
Where to go next
- AI Agents & Automation: news, companies and community questions
- How to evaluate AI Agents & Automation vendors: criteria, questions to ask and red flags
- AI agents: the latest stories and threads on this topic
Written by Techarda editors · Updated . Think a definition is wrong or missing something? Email engage@techarda.com. All terms.