Loading…
Glossary term
Rules and technical controls that keep an AI system within safe and intended behavior.
Guardrails are controls that keep an AI system within acceptable behavior: input filters, output checks, restricted tools, content classifiers and rules in the prompt.
They work in layers, so if one fails another may still catch the problem.
No single safeguard is perfect, so layered guardrails are standard practice for deployed AI.