Guardrails

Limits that stop automation or AI from doing harm, such as approval gates, dry runs, and caps on how much it changes.

Where the book teaches it