OpenAI Details How Codex Sandbox Limits File and Network Access
The built-in `:workspace` profile blocks network access when a Git repository is the workspace root; other profiles set different permissions.

OpenAI detailed how Codex’s sandbox limits command access to files and networks, including a workspace profile that blocks network access when a Git repository is the workspace root.
Permission profiles set the boundaries for a Codex session. Users can choose settings for a task, such as permitting changes in selected directories while disabling network access. For code review, a profile can keep the code under review read-only while allowing documentation lookups.
The built-in :workspace profile has specific defaults when a Git repository is the workspace root. It permits writes within the workspace but keeps .git and .codex read-only. It also allows broad filesystem reads, subject to the user’s existing operating system permissions, and disables network access.
Those restrictions apply to the :workspace profile, not every profile. The selected profile therefore determines which file and network boundaries apply to a session.
On macOS, Linux and Windows, Codex commands and the processes they start inherit the same sandbox boundaries through platform-specific enforcement. This carries the configured limits to child processes launched during a command.
Users and enterprises can customize permissions, and developers can integrate the sandbox into their own tools. The controls let teams configure file and network access for different kinds of work, including editing within a workspace or reviewing code with source files kept read-only.