Codex approval policy untrusted error appears when a current Codex client reads the retired approval_policy = “untrusted” setting. OpenAI states that this value is no longer supported in Codex CLI 0.149.0 and later, desktop 26.818.31338 and later, and matching VS Code extension builds.
The fix is configuration migration, not repeated sign-in or repository deletion. Back up the small configuration file, remove only the obsolete approval-policy value, and keep your separate project trust settings intact.
Table of contents
Codex approval policy untrusted error: confirm the source
Read the full startup message and note the Codex client and version. The error should explicitly mention approval_policy = “untrusted”. If the message instead concerns authentication, usage limits, a repository connection, or sandbox permissions, use the matching diagnostic path rather than editing configuration blindly.
On Windows, the user configuration is normally under %USERPROFILE%.codexconfig.toml. On macOS and Linux it is normally ~/.codex/config.toml. Organization-managed settings, command-line arguments, or named profiles can override that file, so search every active source.
Never post the complete configuration publicly. It can reference private paths, tool servers, profiles, or environment names even when it does not contain passwords.
Update Codex before migrating the setting
Install the current supported Codex build through the same channel you originally used, then restart the terminal, desktop app, or editor. Mixed client versions can make a configuration appear valid in one surface and fail in another.
Record the working version for teammates and CI jobs. The Codex approval policy untrusted error may remain on one machine if a global package, desktop app, or extension was not updated with the others.
Remove only the obsolete approval-policy line
Close Codex, make a copy of config.toml, and remove the line that assigns untrusted to approval_policy. Save valid TOML and restart Codex. Do not delete the entire .codex directory; it can contain unrelated configuration and useful local state.
Check named profile sections in the same file. A top-level correction will not help if the active profile still sets approval_policy to the retired value. Also review startup scripts or aliases that pass –ask-for-approval untrusted.
Choose supported restrictive controls
OpenAI recommends sandbox_mode = “read-only” with approval_policy = “on-request” as a restrictive alternative. Read-only limits writes, while on-request allows the client to ask before an action needs broader access. Confirm the exact behavior in the current Codex documentation for your environment.
The project-level trust_level = “untrusted” setting is separate and remains supported. Do not remove it merely because the names look similar. The Codex approval policy untrusted error concerns the approval_policy value, not project trust classification.
Check managed configuration and active profiles
If the error returns, inspect organization-managed configuration, allowed_approval_policies, the selected profile, editor settings, and launch commands. A managed value may be read after your personal file and reintroduce the unsupported option.
On a work device, send the administrator the exact error, client version, profile name, and sanitized configuration path. Do not bypass mandatory controls. The administrator should deploy a supported policy across managed clients.
Run diagnostics and verify with a harmless task
On Windows, OpenAI documents Codex Doctor for startup, connectivity, and performance diagnostics. Run the supported diagnostic command for your installed client, then start Codex in a test repository and ask it to explain a small read-only file.
Confirm that Codex starts, shows the intended approval behavior, and does not reproduce the message. If Codex approval policy untrusted error persists, collect the version, platform, client surface, sanitized config, active profile, and diagnostic output for OpenAI Support.
Frequently asked questions about Codex approval policy untrusted error
Why did Codex approval policy untrusted error start suddenly?
A newer Codex release may have reached a machine that still had the retired approval_policy value in its configuration or active profile.
Is trust_level = “untrusted” also unsupported?
No. OpenAI distinguishes the separate project-level trust_level setting, which remains supported, from the retired approval_policy value.
Should I delete the .codex folder?
No. Back up and edit the specific configuration source. Deleting the whole folder can remove unrelated settings and does not fix managed overrides.
Official sources and related guides
Use the current vendor documentation for version-specific details. The links below are direct, dofollow references.
Continue with these TrueFixGuides resources:
