OpenAI’s Codex app for Mac just got a major upgrade, and it’s the kind that makes you feel like you’re living in the future. You can now send Codex a task from your phone, and it will go ahead and use apps on your Mac, even if the screen is off and the computer is locked.
One of the major issues with AI agents like Claude Cowork or OpenAI Codex is that they need your Mac to be unlocked and running for them to work. That’s a major security flaw that will prevent most users from using these features. With its latest Codex update, OpenAI has seemingly fixed this issue.
A post on X from OpenAI Developers showed a locked Mac with a “Codex is Using Your Mac” overlay on the display.
How does it actually work?
For this feature to work, you need to install the Computer Use plugin and grant it the necessary permissions to run. You can find the locked computer use setting by going to Codex → Settings → Computer Use.
After you enable it, you can send any task from your phone to your Mac, even when it’s locked. When Codex receives the task, and it needs to access your Mac after it locks, it will temporarily unlock the computer in the background while keeping the screen protected.
If it detects that someone has touched the keyboard or mouse, it will immediately relock the Mac and stop the automatic unlock until you manually log back in. That’s a good safety net to have if your computer sits in a room accessible by other people.
Should you be worried about security?
OpenAI has built in several safeguards. Each unlock attempt is short-lived and scoped only to the active Codex task. Codex also covers every display while your desktop is temporarily unlocked, so no one nearby can sneak a peek at your screen.
You also stay in control throughout the process. Codex asks for permission before operating each new app, and you can stop the task or take over your computer at any time. There’s also an “Always allow” option for apps you fully trust.
As for availability, the feature is not available in the European Economic Area, the UK, or Switzerland at launch. It also cannot automate Terminal apps, Codex itself, or any system-level admin prompts.
OpenAI has built in enough protective measures to satisfy most users. Whether you decide to trust it enough to use it is up to you.