OpenAI quietly shipped one of the more significant capability updates of the year. Codex, its AI coding and task agent, can now operate Mac applications after the screen locks — and you can trigger and monitor those tasks from your phone while you’re away from your desk.

The feature is called Locked Use, and it was announced on May 21, 2026, as part of a broader Codex update that also included improved computer use controls and 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.

What Locked Use Actually Does

The practical implication is straightforward: you can set Codex working on something, lock your Mac, and go to sleep. The agent keeps going.

Previously, agentic computer use on desktop required the machine to remain awake and accessible, which created an obvious limitation for any task that ran longer than a working session. The workaround was leaving your laptop open and screen-on, which is exactly the kind of friction that makes enterprise teams hesitant to trust AI agents with overnight or weekend work.

Codex solves this by installing an Apple authorization plug-in that participates in the macOS unlock flow. When the agent needs to interact with a desktop application, it temporarily unlocks the Mac internally while keeping the screen blocked and local access restricted. Once the task is complete or local input is detected, the Mac relocks automatically.

The safeguards built into this are worth noting:

• Authorization is short-lived and scoped to the active agent task
• Displays are covered so the screen remains effectively locked for anyone in the room
• Any local keyboard or mouse input immediately relocks the machine
• A manual-unlock fallback exists if something goes wrong

This is not a general remote-access tool. Codex cannot unlock your Mac for other apps or local processes. The feature is explicitly limited to active, trusted Codex computer use sessions.

The Constraint Worth Knowing

Locked Use is not available in the European Economic Area, the United Kingdom, or Switzerland at launch, which limits its relevance for enterprise teams operating in those jurisdictions. It also cannot automate Terminal applications, Codex itself, or system-level administrator prompts.

For teams outside those regions, the path is straightforward: install the Computer Use plug-in, grant Screen Recording and Accessibility permissions, and explicitly define which applications Codex is allowed to interact with.

What This Means for Business

The bigger shift here is philosophical. Agentic AI has been sold for two years on the premise that AI agents will work through the night, handle repetitive workflows autonomously, and free up human attention for higher-value decisions. The reality has been messier: agents that stall when a window needs a click, tools that require a logged-in session to stay active, workflows that break the moment a screen saver kicks in.

Locked Use is one of the cleaner solutions to that gap. A developer or operations lead can hand off a task to Codex before logging off and expect it to have progressed by morning.

The use cases that make practical sense today include: reproducing graphical interface bugs in desktop software, running end-to-end test flows that require application navigation, executing repetitive data entry tasks in tools without APIs, and validating visual workflows in business applications that lack automation-friendly backends.

For knowledge workers and technical teams, this shifts the conversation from “can AI agents work autonomously?” to “what should I delegate to one tonight?”

What Enterprise DNA Is Watching

The direction this points toward is clear. As computer use capabilities mature across OpenAI, Anthropic, and Google, the question of whether AI agents can reliably handle real-world desktop workflows shifts from speculation to systems design. Which processes deserve an agent? What governance do you put around unsupervised automation? How do you audit what the agent actually did?

These are the questions organisations need answers to before the technology outpaces their readiness to use it well.

If you are thinking through what AI agent deployment looks like for your business, that is exactly the conversation our Omni Advisory team works through with clients.