Hit rate limit in Codex app, switched to API key, now get:
{ "error": { "message": "The encrypted content gAAA...2A== 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" } }
Context:
Was using ChatGPT auth originally
Switched to API key (personal org)
Session encrypted with old org context
Ask: How can I continue the same conversation?
Tried: Fresh session + manual context paste (lossy for long convos).
It is safe to say: No continuation, unless you go into the local .codex conversation storage and clean up the individual assistant responses and remove reasoning items of output. Then the next modality to fail will be tool calling that happened within reasoning on a hosted product.
I haven’t gone so far in consuming OpenAI’s presentation to even care if ChatGPT is a different base format of storage (but did go into the .rs to see what they are up to over the wire..) Then no Mac to even see if all the closed-source app local data isn’t simply made impenetrable.
The encrypted reasoning that prevents you from seeing the true model internal generation is likely org-specific so that one breach of the algorithm isn’t company-wide revelation.
We can see encrypted reasoning has layers of metadata that can be extracted from a container in a first layer of common encryption or obfuscation to report on the org id mismatch, instead of merely failing.
You can see I posited this earlier as the likely implementation, which your error answers:
Hello folks, I just want to make a point here, that I just created my user here, exactly because of this problem.
So first note, it was really amazing that I stumbled upon this problem, then right after I logged in this forum, I saw this message in the front, the default list. It’s perhaps a coincidence or our OpenAI knows that I stumbled upon the problem, on the other end (and filtered for me).