Safety, Ethics & Governance · intermediate · concept 285 of 311 · Updated Nov 4, 2026

🚧AI Guardrails

Guardrails are checks around a model that block unsafe input or output: content filters, schema validation, permission limits and human approval steps. They exist because a model cannot be relied on to police itself. Strong guardrails limit what the system can do, not just what it says, since filtering text alone will not stop every attack.