AI guardrails are controls, policies, filters, evaluations, or runtime checks that constrain AI system behavior toward acceptable, safe, and governed outcomes.
Teams use guardrails to reduce harmful output, data leakage, policy violations, unsafe tool calls, and other operational risks in AI applications.
Additional Resource: Authoritative source.
Related concepts include Responsible AI.