Guardrails

AI guardrails monitor and control how models and applications process inputs, generate outputs, and perform actions. They can enforce policies related to harmful content, sensitive data, prompt injection, access permissions, prohibited topics, and acceptable AI use. Guardrails may operate before a prompt reaches the model, after a response is generated, or around actions performed by AI agents.

AI systems can produce unpredictable outputs or be manipulated into behaving outside their intended purpose. Guardrails help reduce these risks by establishing enforceable boundaries around AI behavior. They can also help organizations apply security policies consistently while supporting safer deployment of generative and agentic AI.