Failed to authorize remote control
Resolved 💬 41 comments Opened May 14, 2026 by MerijnRaaijmakers Closed May 16, 2026
What version of the Codex App are you using (From “About Codex” dialog)?
Codex 26.513.20950
What subscription do you have?
Pro plan
What platform is your computer?
MacOS - Darwin 25.2.0 arm64 arm
What issue are you seeing?
Codex just got updated and trying to setup the "Control other devices" under the connections tab in settings. Sadly getting the "Failed to authorize remote control" error after pressing the "Authorize on chatgpt.com" button.
What steps can reproduce the bug?
- Go to the connections page under the settings panel.
- Click the Control other devices tab.
- Click "Set up".
- On the popup panel click "Authorize on chatgpt.com"
- Error happens "Failed to authorize remote control"
What is the expected behavior?
Browser opens to authorize on chatgpt.com
Additional information
Already got SSH configured without problems.
41 Comments
I can reproduce this on another Mac with the same Codex app build.
Environment:
26.513.20950, build2816(com.openai.codex)currentVersion=0.131.0-alpha.9codex-cli 0.130.026.3(25D125), arm64, Apple M1 MaxObserved behavior:
Failed to authorize remote control2026-05-14T21:13:36Zand2026-05-14T21:13:45ZLocal log clues, with account/user identifiers redacted:
The actionable detail from the full local log is that the request has matching
headerChatGptAccountId/tokenAccountId, but the enrollment response'sresponseAccountUserIdappears to use the bare user-id shape whileselectedAccountUserId/tokenAccountUserIduse the combineduser_id__account_idshape.This looks like an account identifier normalization/comparison mismatch before the browser authorization page is opened, not just a browser launch failure.
Another repro from a Mac with the same current desktop build.
Environment:
26.513.20950, build2816(com.openai.codex)codex-cli 0.130.026.4.1(25E253), arm64Observed behavior:
Connections -> Control other devices -> Set up -> Authorize on chatgpt.comFailed to authorize remote control.2026-05-14 16:32America/Chicago.This looks aligned with the existing report: the failure appears specific to authorizing this Mac to control other signed-in desktop devices, not general ChatGPT/Codex auth and not mobile remote-control connectivity.
I checked the accessible local logs on this machine, but did not find a clean
remote_control_client_enrollment_start_account_mismatchline to add without including unrelated/sensitive telemetry. The visible behavior and build/version match this issue exactly.Codex Desktop 26.513.20950, build 2816 (com.openai.codex)
macOS 26.3 (25D125), arm64
ChatGPT-authenticated account
Able to connect iPhone with same crews but not M1 controlling an M4 Mac
Mobile/iPhone remote control works for the same account, but Codex M1 Desktop-to-Codex M4 Desktop authorization fails from Settings -> Connections -> Control other devices -> Authorize on chatgpt.com with “Failed to authorize remote control”.
I am seeing the same failure on Codex
26.513.20950(2816) on macOS26.5/ Darwin25.5.0arm64.I dug into the local Codex logs around the toast. The visible failure is preceded by a remote-control enrollment account mismatch, but the values look like a format mismatch rather than a real account mismatch:
This happened at
2026-05-14T23:16:11Zimmediately after clicking Authorize on chatgpt.com. The same pattern appears on repeated attempts.I also inspected the packaged app code locally and it looks like enrollment start does a strict comparison between the API response
account_user_idand the current token account-user id. In this case the ChatGPT account id matches, but the enrollment API returns the bare user id while the token/current account user id includes the__<account-id>suffix. That appears to make Codex reject a valid enrollment response.There is also repeated log noise in the same session where
experimentalFeature/enablement/setforauth_elicitationreturnsunsupported feature enablement, which may indicate client/server feature-contract skew, but the direct cause of the toast is the remote-control enrollment account-id comparison above.Also getting this same issue
I can reproduce the same issue.
Environment:
Authorize on chatgpt.comObserved:
Settings → Connections → Control other devices → Set up → Authorize on chatgpt.comCodex immediately shows:
Failed to authorize remote controlThe browser authorization flow does not open or complete.
I checked the local Codex Desktop logs under:
~/Library/Logs/com.openai.codex/YYYY/MM/DDThe local logs show repeated occurrences of the same pattern:
remote_control_client_enrollment_start_responsefollowed by:
remote_control_client_enrollment_start_account_mismatchfollowed by:
remote_control_authorize_failedwith:
errorMessage="Remote control enrollment start does not match current account."Redacted shape of the relevant fields:
responseAccountUserId=<user-id>but:
selectedAccountUserId=<user-id>__<account-id>and:
tokenAccountUserId=<user-id>__<account-id>while:
headerChatGptAccountId=<account-id>and:
tokenAccountId=<account-id>match.
This appears to be the same account identifier normalization/comparison mismatch described above: the enrollment API response returns the bare user id, while the selected/token account user id uses the combined
user_id__account_idshape. Codex then rejects the enrollment before the browser authorization flow opens.same issue. it seems no user-side workaround so far
Experiencing the same here. Very similar setup to posters above.
Same issue here.
Additionally I started experiencing weird web ChatGPT issues today. Multiple 404 and 403 errors related to shared Project files/images and even billing. Is there some kind of account identity/object auth issue that is related to this?
I’m seeing this too on Codex
26.513.20950on macOS.The UI shows the same toast after clicking Authorize on chatgpt.com:
I checked the local Codex logs and found the underlying failure:
So it looks like the remote-control enrollment response account/user id is not matching the currently selected Codex account context. This happens before the browser authorization page opens.
I have the same issue, after I revoke the chatgpt codex auth
I can reproduce what looks like the same issue, with one extra detail: mobile remote control works, while desktop-to-desktop remote-control authorization fails.
Setup
Two macOS machines, both signed into the same ChatGPT Personal account/workspace:
arhitek, active roots under/Users/arhitek/Code/...dejantomka, active roots under/Users/dejantomka/Code/...Versions on both machines:
Connection matrix
So this does not appear to be a general mobile remote-control failure, and it does not appear to be an SSH remote project issue. It seems specific to desktop-to-desktop remote-control enrollment/account matching.
Failing flow
In Codex Desktop:
Actual result:
The error appears immediately as a red toast inside Codex Desktop. The browser does not open.
Expected result: Codex Desktop should start the remote-control authorization flow and enroll the other desktop host.
SSH remote project status
SSH remote projects work in both directions.
One SSH issue did appear initially: Studio → MBP could not find
codexbecause non-interactive SSH on the MBP did not include/opt/homebrew/bininPATH.I fixed that on the MBP with a minimal
~/.zshenv:After that, this works from Studio:
Output:
After this PATH fix, the Studio → MBP SSH remote project connection updated successfully.
Troubleshooting already tried
Relevant log lines
After reproducing the error, I checked:
Relevant redacted lines:
Key error:
There are also repeated nearby feature sync errors:
and similar variants without
mentions_v2.Question
Could this be a desktop-to-desktop remote-control enrollment bug?
The important distinction is:
Is there a safe local state reset/workaround for this? I would prefer not to delete the entire
~/.codexdirectory because it contains sessions, history, logs, and local Codex state.If a reset is needed, which files are safe to remove/regenerate? For example, is it safe to reset only:
or is there another specific remote-control enrollment/cache file that should be cleared?
I’m seeing the same issue on two Macs.
Plan: ChatGPT Pro
Codex App: 26.513.20950
Codex CLI: 0.130.0
Platform: macOS / Apple Silicon
Both machines show:
codex features list | grep remote_control
remote_control under development false
Codex App shows:
Failed to authorize remote control
Relevant logs:
remote_control_authorize_failed
Remote control enrollment start does not match current account.
remote_control_client_enrollment_start_account_mismatch
This happens after updating Codex, signing out/in, and using the same ChatGPT Pro account on both machines.
Same issue. no way to bypass it
Same. I can remotely control from iPhone to Mac, but cannot control from Mac to Mac. Tested between macbook pro m4, mac studio m2, mac studio m3. I'm on MacOS 26.5 on all.
Eagerly waited for the Connections tab to appear and now also facing the same issue :(
following - having the same issue
same issue here
same issue
same issue
Same issue, can reproduce on 2 machines, mac 2 mac
I’m hitting the same issue on two Macs.
Flow that fails:
Result:
Codex immediately shows “Failed to authorize remote control” and never opens a browser.
Relevant log line:
remote_control_client_enrollment_start_account_mismatchThe logs appear to show Codex gets an enrollment response, then rejects it locally as an account mismatch even though the selected account/workspace is the same on both machines.
same issue here
Same here.
Also, once you set up the mobile device and remove the mobile ACL, there’s no way to reconnect it.
Same issue here. On iPhone I can see the remote environment (Ubuntu using experimental remote-control mode) and chat with it (although with unrelated issues related to authorizations/approvals for actions). When I first set it up, the remote host was showing on my mac but no longer does after a couple minutes and a codex restart
Same issue. Can connect through the ios app but not through another macos codex app. So many bugs.
I can reproduce this as a high-severity blocker on the same ChatGPT account, and I want to add a stronger impact signal here because this is not just a minor remote-connection inconvenience.
This feature worked once for me yesterday: I was able to set up mobile access to a local Codex App host. Shortly afterward it became disconnected, and since then I have been unable to authorize remote control again. The same ChatGPT account now fails with the same behavior on another computer as well, so this does not look like one corrupted local Codex profile.
This is a serious adoption blocker for me. I am trying to teach coworkers and students how to use Codex remote workflows, but the current failure mode makes the feature unusable and very hard to explain or recommend. From a user perspective, the setup flow presents itself as available, then fails with a generic authorization error. There is no recovery path in the UI, no clear way to reset the broken server-side enrollment, and repeated sign-out/sign-in attempts do not fix it.
Environment from one affected Mac:
What I tried:
Current behavior:
The latest local log attempt on this Mac was around:
Redacted log sequence:
The important comparison from the same attempt:
That strongly suggests the enrollment start response is returning the bare auth user id, while the desktop client expects the account-scoped
chatgpt_account_user_idform. The app then rejects the enrollment response even though the account id itself matches and the response appears to refer to the same underlying user.I also checked the local remote-control state after cleanup/re-login. These keys were absent, so there does not appear to be an old local client enrollment to refresh or delete:
At the same time, Codex still refreshes a remote-control environment and reports one disconnected connection:
So from the outside this looks like:
Please prioritize this. Remote control is a core workflow for using Codex across machines and for teaching teams how to adopt Codex. Right now the feature can enter a broken account/enrollment state with no user-visible recovery path and no reliable workaround. If there is a safe server-side reset or a specific local state reset that users should perform, please document it clearly; otherwise this looks like it needs a client/server fix for the account user id comparison or the enrollment response shape.
I've just gotten the Version 26.513.31313 (2867) and this issue seems to be resolved!
the latest update resolved the issue for me Version 26.513.31313 (2867)
Verified the fix in
Version 26.513.31313 (2867)I am still having issues with this to this day. I have no restrictions I can find nor company restrictions (these are my own personal devices) can't find any reason for it
What version of the app are you on and what platform, and same for remote host? also did you try logging out and logging back in completely? @christophervankammen
I am on Codex App 26.519.81530 (3178) MacOS and i am still facing this issue as well today
I'm also having this issue on Codex 26.519.81530 (3178) for macOS.
also having this issue on Codex 26.519.81530 (3178) from macOS, Pro Plan
The same issue Version 26.601.21317
Same for Version 26.602.30954
Solved: Looks like Codex pairing requires MFA. On your iPhone, install Google Authenticator, then go to your ChatGPT/OpenAI account Security settings and turn on Two-Factor Authentication. Scan the QR code with Google Authenticator, save the backup codes, then sign out and back into Codex and try pairing again. It only takes a couple of minutes.
Didnt work, still the same for me even after setting up MFA and relogin
this is still happening on intel macs
Still happening on both Intel and AS Macs