Guardrails

The rules and filters that keep an AI from producing harmful, unsafe, or off-limits output.

Why it matters

Guardrails are what stand between a capable model and misuse, though they are not perfect.