Sometimes, when you open the Codex desktop app, a screen titled "Organization settings could not be loaded" appears and you cannot get any further. The wording points to the network or a proxy, but according to bug reports about Codex on GitHub, the cause is often something other than the network. This article covers a real case on a Windows PC on October 22, 2026 that was fixed, the types of causes users have reported, how to tell them apart in the logs, and what is better not to try first.

First

Press "Retry"

Errors that appear right after an update can sometimes be fixed by retrying or reopening the app. In a real case, a single retry fixed it.

Leave for later

"Sign out"

The cause is often not the account. On Windows, some reports say signing out leads to an error screen.

If it still won't go away

Identify the type from the logs

Right after an update, with WSL, after an abnormal exit, or on the network, the log lines that serve as clues are different.

Table of contents

1. The screen and the English text

In the English interface, the screen appears as follows.

Organization settings could not be loaded
The app is paused until your organization settings can be loaded safely. A network connection, proxy, or firewall may be preventing access. Check your connection and try again, or sign out to use another account. If the problem continues, contact your IT administrator.

Below the message are three buttons: "Retry," "Sign out," and "Quit."

Some reports on GitHub quote a different wording from Organization settings could not be loaded: the title "Unable to load organization settings" with the shorter message "The app is paused until organization settings can be safely loaded. Network, proxy, or firewall may be blocking access."

The screen says "organization settings," but it does not appear only with work or school accounts. Users have reported it with individual Plus accounts and with people using an API key. Each time Codex starts, it checks whether an administrator has set any policies (listed in the logs as managed preferences), and it stops the app until that check finishes. If loading those settings fails or times out, this screen appears regardless of the underlying cause.

Sources: openai/codex #54478 and #53100 (bug reports on GitHub for Codex, written by users).

2. What to do first

  1. Press "Retry"If the error appeared right after an app update, this may get you through.
  2. If that doesn't work, quit the app with "Quit" and reopen itStart Codex again from the taskbar or the Start menu.
  3. If it still appears, check the networkCan you open chatgpt.com in a browser? If you use a company proxy or VPN, try again with it turned off.
  4. If it still appears, look at the cause type in the logsSee chapter 5 for how to tell them apart. The fix differs by type.

Sometimes the problem is on OpenAI's side, so it is also worth checking the OpenAI status page.

3. A real case that was fixed: right after an update

On the morning of October 22, 2026, this screen appeared on a Windows 11 PC used by AI Arte (Codex 26.1016.6573.0, Microsoft Store version). A single press of "Retry" fixed it. The app's log shows the events happened in this order (times are Japan Standard Time).

01:11The Codex app was updated to 26.1016.
06:37The internal engine was replaced, and the app restarted. It began unpacking the new engine into a folder on the PC.
06:40Almost at the same time the unpacking finished, the log recorded "Desktop startup policy read failed (failed to read startup settings) … reason=timeout." This is when the screen appeared.
After thatPressing "Retry" brought up the normal screen.

So in this case, the cause was not the network. It was that the first launch right after the update took a long time to prepare, and loading the settings timed out. At that time, the connection to OpenAI's servers was fast, and no proxy was in use.

GitHub also has reports of the same type. The analysis there is that on the first launch after an update, the process of unpacking the engine and related files runs before the organization settings check, which slows things down. Because the unpacked files are reused from the next launch onward, the second attempt goes through quickly. However, this analysis is based on the reporters' own investigation, and OpenAI has not confirmed it as the cause.

Source: openai/codex #55516 (user report). The real case is from the app log on AI Arte's PC (October 22, 2026).

4. Reported types of causes

Bug reports about Codex on GitHub include many examples of this same screen (as of October 22, 2026, most are still unresolved). All of them are user reports. OpenAI has not yet officially announced a cause or a fix for them.

TypeWhen it happensReported fix
Right after an updateThe first launch after the app or engine was updatedRetry or reopen the app (the example in chapter 3)
WSL settingThe agent is set to run in Windows Subsystem for Linux (Windows)There are reports that the app starts again after switching the execution environment back to Windows
After an abnormal exitAppears every time after the app crashed or stopped during an update (more likely with a large conversation history)In one case where the history database had stopped partway through processing, the reporter fixed it by editing the database directly
Proxy, VPN, or firewallYou are using a company network or a proxy on your own deviceCheck the connection to chatgpt.com, and try again with the proxy or VPN turned off
Mac and LinuxReports with the same title also exist for macOS and LinuxThe causes vary, and there is no fixed solution

For the "after an abnormal exit" type, the reporter fixed the problem by editing the contents of the history database. A mistake could damage your conversation history, so if you want to try it, always back up the entire database file first. If you are not confident, it is safer to add information to the GitHub report and wait for a fix.

Sources: #55516 (right after an update), #54478 (WSL), #52836 (after an abnormal exit), #55207 (proxy), #55245 (macOS), #53100 (Linux). All are bug reports in openai/codex, checked on October 22, 2026.

5. Telling the types apart in the logs

In the Microsoft Store version on Windows, the app's logs are stored by date in the following folder. Paste it into the address bar of File Explorer to open it.

%LOCALAPPDATA%\Packages\OpenAI.Codex_（alphanumeric characters）\LocalCache\Local\Codex\Logs\（year）\（month）\（day）

Open the .log file for the time the error appeared in Notepad, and look for the following lines. Times are written in UTC (Japan time minus 9 hours).

Desktop startup policy read failed … reason=timeoutLoading the settings timed out. If lines for runtime_component_resolved appear just before it and the gap is several minutes, the type is right after an update. Retrying often gets it through.
bundled_executable_relocation_failed (…\bin\wsl\…)Preparing a component to run through WSL failed. This is the WSL type.
state db backfill … timed outProcessing the history database does not finish. This is the after-an-abnormal-exit type. This line does not appear in the app log. The reporter found it by running a Codex component separately.
403 · cloudflare_challengeCommunication is being blocked at OpenAI's entry point. This is common with the different error in chapter 8.

Even without opening the logs, the following two checks can narrow things down quite a bit.

Does ChatGPT work in a browser?If Codex works normally on chatgpt.com in a browser but only the app stops, the network type is unlikely. Suspect right after an update, WSL, or after an abnormal exit.
What happened just before?If the app was updated overnight, it is probably the right-after-an-update type. If you switched the agent's execution environment to WSL, it is probably the WSL type. If the app crashed or was force-quit, it is probably the after-an-abnormal-exit type.

If it is the WSL type, the easiest thing to test is switching the agent's execution environment back to Windows. In #54478, a user reported that the app started after changing "Agent environment" in the app settings from "Windows Subsystem for Linux" back to Windows. Another user confirmed on the same PC several times that turning off this setting (runCodexInWindowsSubsystemForLinux under [desktop]) lets the app start, and turning it back on brings the screen back.

The logs also contain information about the account you are signed in with. When you share them on GitHub or elsewhere, include only the relevant few lines, and remove email addresses and IDs before posting.

6. What not to do first

Signing out right awayThe screen suggests it, but many of the causes are not related to the account. One report says that with the WSL type, signing out returned to the same screen. Another report says that on the Windows version, signing out does not return you to the sign-in screen but shows an error screen, and you cannot sign back in until you fully quit the app.
Deleting the Codex folderIn a thread about a different symptom, where the app would not start, one post says deleting the entire Codex data folder (.codex) fixed it. In the same thread, another post suggests that renaming the folder instead of deleting it keeps important data safe. Rename it or back it up somewhere else rather than deleting it.
Reinstalling right awayIn a Linux report, reinstalling did not fix the problem. Conversation history and data remain after a reinstall, so if the cause is there, the same screen will appear.

If you have moved Codex's data location to another drive, also check that the drive you moved it to is connected. For how to move it and the points to watch out for, see How to move Codex's storage location to another drive on Windows.

Sources: #54478 and #53100 (GitHub bug reports), Sign out on Codex for Windows shows an error screen, and Bug: Codex Desktop App Doesn't Load, Stuck on Logo Screen (OpenAI Community, user posts).

7. Information to include if it still doesn't work

If the problem persists, search the openai/codex Issues on GitHub for reports of the same symptom, and add your information to one of them. (Before opening a new report, check that no similar one already exists.) The following information is useful.

  • The version of the Codex app (shown under "About Codex" in the app) and your OS version
  • The date and time the error started, and what happened just before (an update, an abnormal exit, a settings change, etc.)
  • The agent's execution environment (Windows or WSL), and whether you use a proxy or VPN
  • Any of the lines in chapter 5 that appeared in your log (only those lines)

8. A similar but different error

There is also a different error in which the app opens normally, but when you try to start a new chat, it shows "Couldn't load workspace settings." This one does not stop the app at startup. It blocks starting new chats and sending messages. Many GitHub reports show that communication is blocked with a 403 error at OpenAI's entry point (Cloudflare). Some users report that ChatGPT works in a browser but the app alone stops.

If Codex's built-in browser is not working, see a recovery example for when Codex's browser is unavailable. If you see Docker or WSL errors after moving the app to another drive, see Docker and WSL errors after moving a Windows app.

Sources: #55489 and #55513 (GitHub bug reports).

Summary

  • "Organization settings could not be loaded" appears when loading settings at startup fails or times out. The cause is not always the network.
  • First press "Retry." If that doesn't work, press "Quit" and reopen the app. In the case of an error right after an update, a single retry fixed it.
  • The reported types centre on four causes: right after an update, a WSL execution setting, after an abnormal exit, and a proxy or VPN.
  • You can tell the types apart by the lines in the log. A timeout right after an update points to that type, and a failure in a WSL component points to the WSL type.
  • Don't sign out right away. Before you delete or reinstall anything, back up the Codex folder.

FAQ

Q. Does this appear with a non-work account?

A. Yes. Users have reported it with individual Plus accounts and with people using an API key. "Organization settings" is the name of the process that checks whether an administrator has set any policies.

Q. Will it fix itself if I wait?

A. For the right-after-an-update type, a retry or reopening the app after preparation finishes may get you through. For the WSL type and the after-an-abnormal-exit type, many reports say the same screen returns no matter how many times you reopen the app, so check the log to confirm the type.

Q. Is my conversation history gone?

A. Seeing this screen alone does not delete it. In the case in chapter 3, the history remained as it was after the retry. The only risk of losing it is if you delete files yourself.

Q. Will OpenAI fix this?

A. As of October 22, 2026, many of the related reports remain unresolved on GitHub. No official fix has been published. Adding information to the reports can help move a fix forward.

Sources