AI Glossary termGuardrails
Checks around a model that block bad inputs or outputs - policy filters, schema validation, spend limits.
Also called: safety filters
Checks around a model that block bad inputs or outputs - policy filters, schema validation, spend limits.
Why it mattersAgents especially need them. Anything that can act can act wrongly.
See also Human in the Loop, Prompt Injection
Explore more AI terms
Browse the full glossary for plain-English definitions across models, agents, data, and safety.
Back to all terms