Codex repeatedly ignores instructions and exposes Python/tool output in chat

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

Version 26.908.40834

What subscription do you have?

ChatGPT Plus

What platform is your computer?

Windows 11 Pro Insider Preview, x64, build 26340.9482.ge_prerelease_im.260905-0654

What issue are you seeing?

Feedback ID: no-active-thread-01a09c73-c88b-7a80-9def-6b6ce0136e43

Over the last two days, despite explicit instructions not to show Python code, Base64, technical output, logs, internal calculations, or large text dumps in chat, ChatGPT/Codex has repeatedly started displaying large amounts of Python code and long chunks of document text directly in the conversation.

This has happened across more than 10 different chats, with roughly 30–50 repeated incidents in total.

I have added this prohibition both to Codex instructions and to the relevant project instructions, but the problem continues repeatedly.

I use ChatGPT/Codex for long-form writing, translation, research, file work and project work. The visible technical output floods the conversation and makes normal work extremely difficult.

Python may be used internally, but it must not be shown in the visible chat unless I explicitly ask to see it. The same applies to Base64, logs, technical dumps, internal calculations and long document/article text.

This is not an isolated incident. It has become a persistent cross-chat regression over the last two days.

What steps can reproduce the bug?

Feedback ID: no-active-thread-01a09c73-c88b-7a80-9def-6b6ce0136e43

What is the expected behavior?

Codex should respect the explicit instruction and keep Python, Base64, logs, internal calculations, tool output, and technical processing hidden.

The visible conversation should contain only the requested result, short status messages, and links to generated files unless I explicitly ask to see code or technical output.

Explicit user and project instructions should also apply to “Analyzed” / tool execution blocks shown by the Codex App.

Additional information

This began approximately two days ago and has become persistent across many unrelated chats in the same project.

I have already configured the prohibition in both Codex Instructions and Project Instructions, but the behavior continues.

The problem is especially disruptive for long-form writing, translation, research, and document work because technical output and large document excerpts flood the conversation.

Platform: Windows 11 Pro Insider Preview, x64, build 26340.9482.ge_prerelease_im.260905-0654

Subscription: ChatGPT Plus

I can provide screenshots showing the explicit instruction and the visible Python/tool output appearing immediately afterward. This began approximately two days ago and has become persistent across many unrelated chats in the same project.

I have already configured the prohibition in both Codex Instructions and Project Instructions, but the behavior continues.

The problem is especially disruptive for long-form writing, translation, research, and document work because technical output and large document excerpts flood the conversation.

Platform: Windows 11 Pro Insider Preview, x64, build 26340.9482.ge_prerelease_im.260905-0654

Subscription: ChatGPT Plus

I can provide screenshots showing the explicit instruction and the visible Python/tool output appearing immediately afterward.

Start by reproducing the reported behavior in the Codex App on Windows 11 using the feedback ID and the stated Codex and project instructions. Compare the visible chat and “Analyzed” or tool execution blocks with the expected hidden-output behavior; done means technical output remains hidden unless explicitly requested.