Codex App crashes and exposes internal system instructions in the error message

Open 💬 2 comments Opened Jul 11, 2026 by memurhaberi

What version of the Codex App are you using (From “About Codex” dialog)?

Version 26.707.31428

What subscription do you have?

ChatGPT Plus

What platform is your computer?

Microsoft Windows NT 10.0.26200.0 x64

What issue are you seeing?

The Codex Windows desktop app crashes and displays a very long internal instruction prompt inside the “Most recent error” section.

Error:

An error has occurred
ChatGPT crashed with the following error:

(code=3221225786, signal=null)

The text shown after “Most recent error” contains what appears to be internal system/developer instructions, including response formatting rules, commentary channel instructions, React guidance, and tool behavior.

This content should not be exposed to the user. The error screen also becomes extremely long and difficult to read because the instructions are displayed with escaped newline characters such as \n.

What steps can reproduce the bug?

  1. Open Codex App for Windows.
  2. Open an existing project or start a Codex session.
  3. Submit a coding request and allow Codex to work on the project.
  4. After some time, the ChatGPT/Codex process crashes.
  5. The crash screen appears with error code 3221225786.
  6. Under “Most recent error,” a large amount of internal system instruction text is displayed.

The problem has occurred in Codex App version 26.707.31428.

I did not intentionally stop the process with Ctrl+C or close the terminal. I have attached screenshots showing the full error.

What is the expected behavior?

The Codex App should complete the request without crashing.

If a crash occurs, it should display a concise and useful technical error message or diagnostic identifier. Internal system prompts, developer instructions, hidden configuration, or escaped instruction text should never be exposed in the error interface.

Additional information

_No response_

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗