Skip to content
All insights

Security · 6 min read

A safety checklist for computer-using agents

Giving an agent the ability to operate your computer is a big step. The good news is that a short list of habits covers most of the risk. Use this checklist when you start, and revisit it when you roll agents out to a team.

Before you start a task

A few seconds of preparation prevents most problems.

  • Close windows with unrelated confidential information.
  • Make sure you are signed in to the right account, especially if you manage several.
  • Decide whether the task needs Act or Guide.
  • Know where the stop button and stop shortcut are.

While the agent works

Glance at progress, especially the first few times you run a new task. A good agent shows what it is doing and asks when something is unclear. If anything looks wrong, stop it and adjust the instruction.

Sensitive steps deserve a human

Passwords, payments, approvals, deletions and permission changes should always involve a person. Configure your instructions so the agent prepares these steps and waits for your confirmation.

Least privilege still applies

An agent acting through your screen can do whatever your account can do. That is a good reason to keep your own access appropriate to your role, to use separate accounts for administrative work and to avoid running tasks while signed in with more access than the task needs.

For teams rolling out agents

Agree on simple policies before broad rollout.

  • Which systems are suitable for Act, and which should use Guide.
  • Which actions always require a confirmation.
  • How to report unexpected behavior.
  • How to disconnect devices when someone changes role or leaves.

Review and improve

Look back at completed tasks regularly. Most issues come from ambiguous instructions rather than the agent itself, and a small clarification usually fixes them for good.

Your move

Try it on your own screen.

Download Odin, press your shortcut and put this guide into practice in minutes.