Bug: Codex cannot resume a section after switching login methods (Plus → API Key)

I can nonfirm same behavior on:
codex-cli 0.38.0 with WSL(Ubuntu 24.x)

Also i have been using chat gpt plus account with codex-cli and have reached limit. Than I have switched to API key and after trying to resume session i got the same error as damtt.

Same exact setup and issue here as @xylik

I can confirm this with Windows 11 Pro / Ubuntu 24 WSL2 and Codex CLI 0.39.0.

For me it is switching from API key to chatGPT Plus.

After upgrading to 0.40 I could finally login using chatGPT Plus. Was able to complete a code review without problem. Then this issue happened when asking a follow-up regarding the CR.

I did start with 'codex resume' and picked a session ended few days back.

Windows 11 Pro / Ubuntu 24 WSL2 / Codex CLI 0.40

I had the same issue on MacBook Air M1, using the VSCode Codex extension (version 0.4.14).

Initially, I was using GPT Plus.

When the weekly limit was reached, I tried logging in using an API Key and also using a different Plus account.

Both attempts resulted in the same error:

stream error: stream disconnected before completion:
You exceeded your current quota, please check your plan and billing details.
For more information on this error, read the docs:
https://platform.openai.com/docs/guides/error-codes/api-errors

Suggestion for Plus users:
If your limit is reached while using the Codex extension, you can still continue using the Codex web version without quota issues.

It seems the problem comes from sharing the same conversation between different accounts or between the API and an account. If I start a new session from API key, that won't be a problem. Please fix this. I wanna sharing the same conversation.

It works fine on 0.42

I am seeing this problem in 0.45

Problem still exists in 0.46.0

+1 It's really frustrating this issue

It also happens when I switch between API keys.

today happening when switching FROM api key to normal account

This issue is very critical, it can't let me use codex normally, I don't know why it happened, I don't even do anything, it's like it's run for a small while, and it failed with that error
fix this as soon as possible

Same problem here, happened just now

the same

same problem, just started happening to me

I'm being affected by the same issue now

same prob just now. Is it the server problem that some people get this prob at one time?

same

the same

Something broke on the OpenAI side of things; I tested different keys and tools; same response everywhere

it was working since last 0200 am so idk

Something broke on the OpenAI side of things; I tested different keys and tools; same response everywhere

<img width="723" height="644" alt="Image" src="https://github.com/user-attachments/assets/7f878f86-f404-455b-a5c4-b055e02ce9e2" />

just gotta wait it out :D

Thanks for reporting. We're aware of the problem, and the team is actively investigating. We're using #8120 to track this problem. I'll post updates there as they become available.

Its not working for me as well:

{
  "error": {
    "message": "The encrypted content gAAA...fQ== could not be verified.",
    "type": "invalid_request_error",
    "param": null,
    "code": "invalid_encrypted_content"
  }
}

It seems more stable now; the error no longer occurs frequently
[update EST 15:31] still have error

Facing the same issue!

can users facing this issue get some kind of compensation?

■ {
"error": {
"message": "The encrypted content gAAA...iD0= could not be verified.",
"type": "invalid_request_error",
"param": null,
"code": "invalid_encrypted_content"
}
}
Im now updating maybe fix is already there 0.73

Seeing it now 0.73.0. Starts happening after a few interactions:

{
"error": {
"message": "The encrypted content gAAA...sKk= could not be verified.",
"type": "invalid_request_error",
"param": null,
"code": "invalid_encrypted_content"
}
}

■ {
"error": {
"message": "The encrypted content gAAA...VQ== could not be verified.",
"type": "invalid_request_error",
"param": null,
"code": "invalid_encrypted_content"
}
}

There is an active incident. It's a server-side issue, so upgrading or downgrading the client will not resolve it. The codex team is working on a fix. I'll post updates to #8120 as they become available.

Already working for me!

it still doesn't work for me

<img width="324" height="319" alt="Image" src="https://github.com/user-attachments/assets/8595fe63-7ad9-4576-9c22-54d144bec695" />

Same error here but happy to hear that they are already working on it :trophy:

<img width="491" height="71" alt="Image" src="https://github.com/user-attachments/assets/5523d6f1-aa98-402b-8c8f-08ef9ebc678a" />

Let's have a break!

It started working for me now

same here, the issue at https://github.com/openai/codex/issues/8120 is marked as resolved, but I am still encountering the same error during use. In addition, newly opened conversations also experience the same issue.

ID 019b2a35-2d0e-79b0-b01d-2bde64d4a28c

This is still an issue, even though it is consistently marked as 'fixed' by @etraut-openai and others.

@alc1986, I haven't marked this as "fixed". I've closed other bug reports as duplicates of this one.

Codex was not designed to allow switching between model providers or login methods within the same session. Supporting this presents some difficult technical hurdles.

We are considering whether to disallow this completely (that is, emit an error if you attempt to resume a session with a different model, model provider, or login method). Alternatively, we may be able to support this in certain limited situations.

We recommend that you avoid using a session across model providers or login methods. If you switch login methods, start a new session.

@etraut-openai It has been marked as resolved numerous times:

• https://github.com/openai/codex/issues/4184
• https://github.com/openai/codex/issues/8120
• https://github.com/openai/codex/issues/8100

Switching from subscription to API always used to work flawlessly. Blocking users from being able to do this doesn't seem like the most sensible fix for the issue. It means that when users hit usage limits they are no longer able to continue working - and spending more money with OpenAI via the API. Instead, it forces them to stop working or switch to Claude or Gemini.

@alc1986, we recently add the ability to purchase more credits for your subscription. You can use these credits to extend your usage without upgrading to a more expensive subscription plan or switching to a different login method. Does that meet your needs?

@etraut-openai Please combine the credits and API topics. It can't be right that you have API credit with OpenAI, that you have credit balance, and both pots expire after 12 months, but you can't use them for similar scenarios. That's just not well thought out.

Alternatively, it should be possible to switch smoothly from login to API method or to activate this fallback automatically.

@alc1986, we recently add the ability to purchase more credits for your subscription. You can use these credits to extend your usage without upgrading to a more expensive subscription plan or switching to a different login method. Does that meet your needs?

@etraut-openai No, unfortunately not. I still have ongoing sessions I paid for using the API, which I am now not able to transfer over to my subscription. Why on earth are you locking people out of their own work? I pay for the usage regardless. This was reported as a bug in December and marked as fixed. It clearly isn't fixed, and now you're saying it's an intended feature. I see no good reason to stop users switching between the API and subscription model at will.

I just got the same error when switching from API -> plan.

What is the status of this?

What is the status of this?

It seems to have been silently fixed.

I'm still getting this error with the latest codex-cli 0.101.0 when switching from ChatGPT login -> API key auth and then trying to resume a session.

■ Error running remote compact task: {
  "error": {
    "message": "The encrypted content gAAA...BA== could not be verified. Reason: Encrypted content
organization_id did not match the target organization.",
    "type": "invalid_request_error",
    "param": null,
    "code": "invalid_encrypted_content"
  }
}

When I reach the rate limit on my ChatGPT subscription, I use Azure OpenAI API instead.
Although no errors occur, the mutual sessions are not displayed at all via /resume in the first place.

• codex-cli 0.104.0

Any update on this?

Encrypted content organization_id did not match the target organization.", "type": "invalid_request_error", "param": null, "code": "invalid_encrypted_content"

This should be fixed in the next Responses API deploy! 🎉