Codex Remote Mac Control: A New Always‑On AI Assistant
OpenAI Codex’s Mac app has gained a powerful new trick: you can now send tasks from your phone and let Codex operate apps on your Mac even while the machine is locked and the screen is off. Instead of needing your desktop unlocked and sitting idle, Codex can briefly unlock macOS in the background to perform remote task execution, then hand control back. An on‑screen overlay signals when “Codex is Using Your Mac,” making this feel less like a hidden process and more like a visible, managed session. For anyone exploring AI mobile automation, this turns Codex into a kind of always‑on AI co‑worker living on your Mac but reachable from your pocket. It narrows the gap between cloud agents and local desktop automation, creating a bridge where OpenAI Codex mobile requests can directly drive native Mac apps without you being physically present.
How Remote Task Execution Works When Your Mac Is Locked
To enable this mode, you first install the Computer Use plugin inside Codex and grant the necessary permissions. In Codex → Settings → Computer Use, there is a specific option that governs locked-computer behavior. Once toggled on, you can issue instructions from your phone, and when Codex needs to touch the Mac after it has locked, it temporarily unlocks the system in the background while keeping the display visually shielded. Every unlock episode is scoped to the active task, so Codex isn’t just roaming freely across your desktop. If Codex detects any keyboard or mouse input, it immediately relocks the Mac and stops auto‑unlocking until you sign in again. This balance allows AI-driven remote task execution while still treating human interaction as the ultimate override, preventing unattended sessions from silently continuing if someone starts using the machine.
New Mobile‑First AI Automation Patterns for Power Users
Because Codex remote Mac control no longer depends on an unlocked desktop, it opens up new AI mobile automation workflows. Developers can trigger local builds, run development tools, or orchestrate multi‑app workflows from their phone without leaving the sofa or the subway. Creators might have Codex assemble documents, manipulate media in desktop apps, or prepare complex spreadsheets while they are away from the keyboard. Power users can treat the Mac as an AI‑driven automation hub that reacts to natural‑language commands sent from OpenAI Codex mobile, instead of maintaining separate scripts or remote‑desktop setups. The key difference is that Codex can now safely act like a background operator, using whatever Mac apps you authorize while you focus elsewhere. It turns the traditional “sit in front of the Mac” productivity model into something closer to server‑style orchestration, but powered by conversational instructions instead of shell scripts.
Security Safeguards and Limitations You Should Understand
This new capability inevitably raises security concerns, but OpenAI has added several guardrails. Each unlock attempt is short-lived and tied explicitly to the active Codex task, reducing the window for misuse. While the Mac is temporarily unlocked, Codex covers all displays so passersby cannot see the desktop, even though apps are running. Codex also requests permission before operating each new app, with an optional “Always allow” setting for tools you fully trust. You can cancel a task or manually reclaim control at any time, and any physical input relocks the machine. There are also hard limits: Codex cannot automate Terminal apps, Codex itself, or system‑level admin prompts, reducing the risk of deep system changes. The result is not zero‑risk remote access, but a constrained model where users retain clear control over what AI can do, when, and inside which applications.