On May 21 to 22, OpenAI officially announced the new Codex desktop agent feature called "Locked Use" (Screen Lock Operation). This feature breaks the traditional restriction that AI agents must work in an "unlocked and screen-on" state, allowing developers to control local applications remotely even when the Mac is locked or the screen is off.

1. Core Breakthrough: No Longer Need to "Monitor" Your Computer

Previously, for long-duration complex engineering tasks (such as large code repository compilation or GUI automation testing), developers often had to use "Caffeine" plugins or external dummy monitors to force the Mac to stay awake.

The release of Locked Use has changed this situation:

  • Background Automation: Users can send task instructions from their phones, and Codex can automatically perform UI operations, adjust settings, or run desktop application workflows while the Mac is locked.

  • Improved Development Experience: This feature greatly enhances the feasibility of "multi-hour tasks," eliminating the need for developers to monitor devices to keep them awake.

2. Technical Implementation and Security Mechanisms

To achieve this functionality while ensuring system security, OpenAI has implemented a strict permission and constraint framework:

  • Apple-Approved Plugin: This feature is implemented through a specific Apple-approved plugin. It is not a general remote unlocking path but allows Codex agents to perform specific tasks under restricted behavioral constraints.

  • Strict Permission Management: To run this feature, users must explicitly grant Screen Recording and Accessibility permissions.

  • Behavioral "Boundaries":

    • Restricted Areas: This feature clearly limits the agent's control over the system's underlying functions, such as being unable to automate terminal (Terminal) operations or Codex itself, effectively preventing malicious use or loss of control.

    • Geographic Restrictions: Considering regional laws and AI agent regulatory policies, this feature initially does not apply to the European Economic Area (EEA), the UK, and Switzerland.

    • Operational Constraints: If the MacBook is closed (triggering a specific sleep mode), this feature will no longer function.

3. Industry Perspective: A New Challenge for Desktop Agent Security

The launch of this Codex feature marks the transition of desktop-level AI agents from "cloud collaboration" to "deep system integration." Industry experts point out that this also raises new considerations for enterprise-level terminal security:

  1. Compliance Challenges: This type of "privileged agent" capable of operating beyond the lock screen poses higher audit and monitoring requirements for enterprise information security management (especially for development machines involving production credentials).

  2. Upgraded Trust Model: How the operating system sets more refined permission levels for such "trusted agents" will become a key focus in future macOS and other desktop system updates.

  3. Security Audit Demands: Currently, developers cannot fully "black-box" this process. In the future, how to provide operation log tracking or session monitoring will be a critical barrier for enterprises deploying these AI tools on a large scale.

4. Function Scope

  • Supported Platforms: Limited to macOS devices with the Codex desktop agent installed.

  • Configuration Method: Users can find the "Locked Use" option in the Codex settings to enable it.

  • Usage Suggestions: Primarily aimed at reproducing GUI interface bugs, automating desktop application configuration adjustments, and handling offline tasks without manual intervention.

OpenAI's latest update not only further solidifies its leading position in the AI-assisted development field, but also explores a high-standard, strictly regulated path for the future of "seamless" AI office experiences.