Dashboard / Courses / Keeping AI Agents Safe / A Pre-Flight Checklist, and What to Do When It Goes Wrong

A Pre-Flight Checklist, and What to Do When It Goes Wrong

Before you switch on any agent or connector, take two minutes with this checklist.

Is it running on a practice setup or copy, not your only version of anything? Does it have the least access that does the job? Are credentials kept out of its reach? Is approval required for anything that sends, deletes, pays or publishes? Is there a spending limit? Do you know where to see what it did, and how to switch it off?

Start small. Give it a task with a low cost of failure, check the result, and only then give it something that matters. Keep a record of what you have connected so you can review it every month and remove what you no longer use.

If something goes wrong, act in this order. Stop the agent. Revoke its access: disconnect the connector, delete or rotate the key, sign out its sessions. Look at the activity log to see what actually happened. Restore from your backup. Then, once things are calm, work out which safety habit would have caught it, and add that to your checklist.

You will make mistakes with agents. Everybody does. The aim is to make them small, visible and reversible.

Takeaway: run the checklist before you switch anything on, start with low-stakes tasks, and if something goes wrong, stop it, revoke access, check the log, and restore.

Tools, prices and features in this area change quickly. Last reviewed September 2026.