[ChatGPT Projects] Large Projects stuck on “Searching project context” with repeated HTTP 429s after Sep 3 service incident

What issue are you seeing?

This issue affects ChatGPT Web → Projects, not the Codex CLI or IDE extension.

Since the ChatGPT/Codex service incident on September 3, 2026, one of my large ChatGPT Projects has become almost unusable.

The Project was working normally before the incident. The problem started afterward and is still reproducible on September 4.

I am not claiming that the September 3 incident definitively caused this issue, but the timing may be relevant because the Project worked normally before it.

When I send even a simple message such as "hi" in a new conversation inside the affected Project, ChatGPT remains stuck on:

"Searching project context"

for several minutes and eventually fails.

During the failure, Chrome DevTools Network shows repeated HTTP 429 responses.

I also captured these backend errors in the Console:

GET https://chatgpt.com/backend-api/tasks
→ 500 Internal Server Error

POST https://chatgpt.com/backend-api/conversation/init
→ 404 Not Found

At some points, after repeated troubleshooting attempts, I also received "Too many requests" while trying to access normal conversations outside Projects.

However, the original Project context-loading problem existed before the repeated testing and before those general "Too many requests" messages appeared.

Other requests during the same affected session continue returning HTTP 200/202, so this is not a complete connectivity failure.

What steps can reproduce the bug?

1. Open chatgpt.com using the affected ChatGPT Plus account.

2. Open an existing large ChatGPT Project containing many previous conversations.

3. Create a completely new conversation inside that Project.

4. Send a very simple message such as:
   "hi"

5. The UI enters "Searching project context" and remains there for several minutes.

6. The request eventually fails or remains stuck.

7. Open Chrome DevTools → Network while reproducing the issue.

8. Multiple fetch requests return HTTP 429 while the Project is attempting to retrieve context.

Troubleshooting and isolation tests already performed:

• Tested in normal Chrome session.
• Tested in Chrome Incognito.
• Cleared ChatGPT site data.
• Tested on mobile.
• Tested from another device.
• Created a brand-new empty Project: it works normally.
• Created another brand-new Project and moved all conversations from the affected Project into it: the same problem returned.

The last test appears particularly relevant.

An empty new Project works normally. After the existing conversation set is moved into that Project, the context-loading failure persists.

This suggests that the behavior may be associated with the existing conversation/context dataset rather than only the original Project ID or Project container.

What is the expected behavior?

Existing ChatGPT Projects should retrieve their Project context and respond normally, as they did before September 3, 2026.

Creating a new conversation inside a large existing Project and sending a simple message should not cause several minutes of "Searching project context", repeated HTTP 429 responses, or eventual request failure.

Moving existing conversations into a newly created Project should also not make that new Project become unusable.

Additional information

Environment:

• ChatGPT Plus
• ChatGPT Web
• Windows 11
• Google Chrome 152
• Also reproduced on mobile
• Brazil / Brasília timezone (UTC-3)
• First observed: September 3, 2026, after the ChatGPT/Codex service incident
• Still reproducible: September 4, 2026

Important isolation results:

• New empty Project → works normally.
• Existing large Project → fails during Project context retrieval.
• New Project after migrating all affected conversations into it → same failure persists.
• Normal non-Project chats generally work.
• The issue reproduces across browser sessions and devices.

Observed network/backend errors:

• Repeated HTTP 429 responses during "Searching project context".
• GET /backend-api/tasks → HTTP 500 Internal Server Error.
• POST /backend-api/conversation/init → HTTP 404 Not Found.

I have already opened a case with OpenAI Support and provided the same troubleshooting information.

Because the issue started after the September 3 service incident and an empty Project works until the existing conversation set is migrated into it, this may involve Project context indexing/retrieval or an abnormal backend/rate-limit state associated with the existing Project conversation dataset.

I am reporting this publicly to determine whether this is a regression affecting other users with large/existing ChatGPT Projects after the September 3 incident.

I do not want to delete the affected conversations or manually rebuild the Project because it contains a large amount of active work.

The report identifies ChatGPT Web Project context retrieval, Chrome DevTools Network, and the /backend-api/tasks and /backend-api/conversation/init endpoints, but names no Codex file, test, or entry point. Start by confirming whether this repository owns those endpoints; done would require a reproducible repository-scoped fix and regression test if ownership is established.