Guardrails
Rules and filters built around an AI system to keep its behavior safe, on-topic, and within intended limits.
01 Definition
Guardrails are the safety measures — content filters, topic restrictions, output validation — layered around an AI model to prevent harmful, off-topic, or unwanted responses, whether built into the model itself or added by the company deploying it.
02 Why It Matters
This is why some AI tools refuse certain requests or stay narrowly focused on one topic — those refusals are usually guardrails working as designed, not the AI 'being difficult.'
03 Example
A company's support chatbot might have guardrails that stop it from discussing competitors or giving legal advice.