Rules, filters, or constraints applied to AI systems to keep their behavior within safe and appropriate bounds.
Friendly Description: Guardrails in AI are like the bumpers in a bowling alley for kids. They don't change the game, but they keep things from going off the rails. Guardrails are rules and filters that help an AI stay safe, on-topic, and respectful, like making sure it doesn't share private info or wander into harmful territory.
Example: A children's tutoring app built on top of an AI might use guardrails to ensure the assistant only discusses school subjects, never uses inappropriate language, and politely redirects if a student asks about something off-topic.