From On-Screen Helper to Always-On AI Desktop Control
OpenAI’s Codex Computer Use feature is quietly shifting from a purely interactive, on-screen assistant into an always-on controller for desktop environments. Initially, Computer Use required a fully unlocked and awake Mac session so the agent could see the screen, move the cursor, and type into applications. Now OpenAI is working on letting Codex operate macOS apps even when a laptop is locked or asleep, extending the capabilities of the remote control feature added to the ChatGPT mobile app in mid-May. That mobile update already allowed iPhone and Android users to review outputs, approve commands, switch models, and send tasks to a Mac running the Codex desktop app. Removing the need for an unlocked session would turn Codex into a background operator that keeps working long after the user closes the lid or walks away from their desk.
Remote Device Automation Without SSH or Manual Logins
The emerging capabilities signal a new model of remote device automation: Codex could control other desktop devices running the Codex app, without requiring SSH access or manual logins. OpenAI is exploring a UI that lets users connect to and operate many machines directly, such as installing Codex on a Mac Mini and orchestrating it from a primary device. Crucially, lifting the unlocked-session restriction would allow a phone to ask Codex to open a desktop app, run through a simulator, test a graphical build, or touch a local data source even while the target laptop is locked or asleep. This bridges a major usability gap in current AI desktop control tools and narrows the functional distance between Codex and Anthropic’s Claude Code remote-control offering, which still runs into the same locked-screen limitations on macOS.
What Locked Device Access Means for Everyday Workflows
If Codex can keep acting inside a locked or sleeping session, everyday workflows could become far more automated and less tethered to physical desks. Developers might trigger GUI tests or simulator runs from their phone while away from the office, while knowledge workers could have Codex prepare environments, open applications, or pre-run repetitive tasks before they even return to their machines. Because this AI desktop control no longer depends on the user being actively logged in, it moves closer to a server-like automation layer living on personal hardware. The shift also alters expectations about what a “locked” device really means: locked would no longer imply idle or inaccessible, but instead “busy” with AI-driven tasks. That convenience, however, comes with a parallel need to rethink who or what is allowed to act on a machine that appears, to the user, to be safely secured.
Security, Governance, and How Platforms May Respond
Extending Codex Computer Use into locked and sleeping sessions raises obvious security and governance questions. macOS has long treated a locked screen as a clear boundary between an active user session and an idle, protected state. Any approach that keeps a screen-driving agent alive behind that lock screen is likely to attract scrutiny from platform security teams, not least because it challenges user expectations about access control. For enterprises, this new class of remote device automation demands stronger identity verification, detailed audit logs, granular permissions, and clear policies defining which tasks Codex can run unattended. Security teams will need visibility into how commands are approved from mobile devices, how access to multiple Codex-enabled desktops is managed, and how to revoke or limit AI desktop control in sensitive environments without sacrificing the productivity benefits that locked device access promises.