Guardrails
Level 2
Rules and checks that constrain what an AI system may say or do.
Term 5 of 6 in Building AI systems
In plain language
Guardrails sit around the model: blocking certain topics, validating the format of an answer, requiring approval before actions. They are ordinary software, not intelligence.
Think of it like this
The bumpers on a bowling lane.
Why it matters
Guardrails are how a probabilistic model becomes safe enough to put in front of customers.