OpenAI's Codex Can Now Unlock Your Mac and Work While You Sleep

Codex's new Locked Computer Use lets your AI agent keep working on your Mac apps from your phone, even after the screen goes dark

·
·
·
OpenAI's Codex Can Now Unlock Your Mac and Work While You Sleep
  • Locked Computer Use: Codex can now operate macOS apps while your screen is off and locked, triggered remotely from your iPhone.
  • Apple auth plug-in: Works via a native macOS authorization plugin that temporarily unlocks the machine only during active, trusted Codex turns.
  • Security guardrails: Display stays covered during unlock, relocks instantly on any local input, and cannot automate Terminal or Codex itself.
  • Part of a bigger drop: Ships alongside Goal Mode GA and Appshots (double-Command to inject any window into a Codex thread).
  • Geographic restrictions: Unavailable at launch in the EEA, UK, and Switzerland, likely due to regulatory concerns around autonomous agents.
  • Pricing: Included in existing ChatGPT plans (Free through Enterprise); requires Codex app version 26.519 or later.

OpenAI's Codex just crossed a threshold that most AI agent tools haven't touched: it can now operate your Mac applications while the screen is off and the machine is locked. You kick off a task, walk away, and Codex keeps going. From your phone.

The feature is called Locked Use, and it was announced as part of a broader Codex update that also included a new feature called Appshots for instant window context. Goal Mode, OpenAI's autonomous multi-step task execution capability, graduated to general availability in the same release. But Locked Computer Use is the one that's turning heads.

The problem it actually solves

If you've ever run a long AI agent task on your Mac, you've hit this wall. Leave your Mac unattended during a multi-hour coding task, and the screen locks, abruptly halting whatever the agent was doing. The workarounds were embarrassing for a 2026 developer stack:

  • Common workarounds included caffeinate to prevent idle sleep, third-party utilities that disabled lock-screen triggers, and even hardware display dongles that fooled the Mac into thinking an external monitor was attached.

OpenAI announced a new capability in its Codex AI assistant that makes it even easier: Codex now has the ability to keep running after your Mac is locked. Developers can kick off a task, lock their Mac, and monitor or review progress remotely from a phone, making multi-hour agentic workflows practical without babysitting a machine.

How it actually works under the hood

This is not a hack or a screen-sharing workaround. According to OpenAI's developer documentation, Locked Use integrates directly with macOS's trusted security architecture. When you enable the toggle in Codex settings, the app installs an Apple authorization plug-in that participates in the standard macOS unlock flow.

When a Codex task accesses an app via Computer Use after your Mac locks, Codex temporarily unlocks the Mac while blocking local use and preserving the locked screen protections. Before unlocking, Codex checks whether the unlock attempt is for an active, trusted computer use turn. Outside that short-lived window, Codex denies the unlock and asks you to unlock manually if needed.

The security constraints are tight by design:

  • The feature works through an Apple authorization plug-in that temporarily unlocks the Mac only during active, trusted Computer Use turns, with safeguards including short-lived authorization, covered displays, and an instant relock the moment local keyboard or pointer input is detected.
  • Every display is covered while the Mac is temporarily unlocked for Codex, so nothing is visible to anyone nearby.
  • The authorization window is short-lived and scoped to the current unlock attempt. Automatic unlock is available only to Codex during active computer use turns.

What it can and can't do

Locked Use inherits all the capabilities of Codex's existing Computer Use feature, which lets the agent see and operate graphical interfaces on macOS. That means clicking through menus, typing, navigating windows, and interacting with the clipboard, all in apps you've explicitly approved. But there are real limits:

  • Codex cannot automate Terminal or itself under Locked Use, limiting but not eliminating local system exposure.
  • It cannot authenticate as an administrator or approve system-level security prompts.
  • It's unavailable in the European Economic Area, the United Kingdom, and Switzerland at launch.

The geographic exclusion of EEA, UK, and Switzerland signals that OpenAI is anticipating regulatory friction around autonomous agents with local system access, a pattern that will shape how AI agent features get scoped and launched globally.

The bigger picture: "Codex Thursday"

OpenAI shipped three Codex updates internally dubbed "Codex Thursday." Goal Mode exits experimental and goes stable across app, IDE extension, and CLI. Appshots lands on macOS, letting you inject the frontmost window into any Codex thread with a double Command key press. And Codex can now keep running on your Mac after the screen locks, including via remote trigger from your phone.

Press both Command keys and Codex grabs the frontmost window: a screenshot plus whatever text it can pull through accessibility. No copying, no pasting, no describing what's on screen. Use goal mode to give Codex a specific milestone, and it will keep working until it gets there, even across hours or days. You can check in and steer, and even pause Codex along the way.

Together they point in one direction: Codex as a background process, not a foreground tool. Appshots reduces the overhead of handing off context. Goal Mode gives the agent a persistent task to work toward. Locked access means the screen does not need to be on for work to happen.

The security conversation

The reaction online has been a mix of awe and unease. The announcement post quickly racked up 1.8 million views, with prominent developer voices like Theo Browne remarking: "At this point I think OpenAI has more MacOS engineers than Apple does." But others are more cautious.

Giving an AI agent the power to unlock your device from a remote order creates a sizable new attack surface, even though OpenAI stresses that the unlock window is brief and limited to the active Codex job. An attacker could gain remote access to your desktop environment without requiring your password or physical presence if your OpenAI account or the Codex agent itself are compromised.

Regarding locked computer use, while it is perceived as "convenient,

Trending
  • No trending articles

Comments

avatar

Next Reads