OpenAI Codex can now control your Mac even when it's locked, accept tasks from your phone

OpenAI Codex Introduces Locked Use Feature for MacBooks

OpenAI has rolled out a significant update to its Codex app for Mac that addresses one of the most persistent challenges facing AI agents: the need for computers to remain unlocked during task execution. The new capability allows OpenAI Codex to run on locked MacBooks, fundamentally changing how developers and users interact with the AI assistant 1

1

.

Previously, AI agents would stop working when a Mac entered sleep mode, forcing developers to employ workarounds like invoking sessions, using utilities to bypass Lock Screen settings, or even installing dummy display dongles to trick the Mac into thinking a display remained attached. These makeshift solutions are now unnecessary with Codex's locked use feature, which can be activated directly in the app's settings 1

1

.

How to Send Tasks From Your Phone to a Locked Mac

The functionality extends beyond simply keeping Codex active on a locked machine. Users can now send tasks from their phone to their Mac, and the AI assistant will execute them even when the screen is off and the computer is locked 2

2

. This remote capability transforms Codex into a truly flexible automation tool that doesn't require constant physical access to the computer.

To enable this feature, users must install the Computer Use plugin and grant necessary app permissions. The setting is accessible through Codex → Settings → Computer Use. Once activated, Codex can unlock the computer in the background while keeping displays protected when it needs to access the Mac after it locks 2

2

.

Security Features and Safeguards Built Into Temporary Unlocking

OpenAI has implemented multiple safeguards to address potential security concerns. According to the company's documentation, the locked use feature is "intentionally narrow" and not designed as a general-purpose remote-unlock path for the Mac 1

1

. The authorization window is short-lived and scoped only to active Codex tasks, with availability limited exclusively to the Codex app for Mac 1

1

.

When Codex temporarily unlocks a Mac, it covers every connected display to prevent anyone nearby from viewing the screen. If the system detects keyboard or mouse activity, it immediately relocks the Mac and stops automatic unlock attempts until the user manually logs back in 2

2

. This provides a critical safety net for computers in shared spaces.

User Control and Permission Management

User control remains central to the feature's design. Codex requests permission before operating each new app, and users can stop tasks or take over their computer at any time. An "Always allow" option exists for apps users fully trust, streamlining the automation process for frequently used applications 2

2

.

The feature does have limitations. It won't work when a MacBook's display is physically shut, as this triggers a different sleep mode 1

1

. Additionally, Codex cannot automate Terminal apps, itself, or any system-level admin prompts 2

2

.

Geographical Restrictions and Availability

The feature faces geographical restrictions at launch. It is not available in the European Economic Area, the UK, or Switzerland 2

2

. These limitations likely reflect ongoing regulatory considerations around AI automation and data privacy in these regions.

For developers and power users who have long struggled with the constraint of keeping their Macs awake during lengthy AI agent tasks, this update represents a practical solution. The balance between automation convenience and security will determine whether mainstream users adopt the feature, but OpenAI's implementation suggests careful consideration of the risks involved in granting AI agents this level of system access.