Guardrails: definition

Guardrails are the rules and checks that keep an AI system inside safe, correct behaviour, blocking off-topic, unsafe, or out-of-scope responses before they reach the user.

Guardrails are the difference between a demo and something you can put in front of customers. They cover input checks (is this even in scope), output checks (did the model stay on-policy), and hard stops (never give medical, legal, or financial advice, always hand off on these triggers).

We treat them as first-class, not an afterthought: every system we ship has explicit rules for what it refuses and where it escalates.

Related terms

System prompt: A system prompt is the standing instruction given to a model before the conversation starts, setting its role, rules, tone, and the boundaries it must keep across every reply.
Human in the loop: Human in the loop is a design where an AI system does the work but a person reviews, approves, or handles the cases the system is unsure about, keeping judgement and accountability with a human.
Hallucination: A hallucination is when a model states something false or invented as if it were fact, because it predicts plausible-sounding text rather than looking anything up.