If an AI system starts behaving in ways you did not expect, who can actually stop it? Here's a practical guide to designing AI shutdown capability, from authority and access controls to safe fallbacks and a realistic kill-switch checklist.