Codex App incorrectly labels a fresh image-only or multimodal turn as "Steered conversation"

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

Version 26.519.41501 (3044)

What subscription do you have?

ChatGPT Enterprise

What platform is your computer?

Darwin 25.5.0 arm64 arm

What issue are you seeing?

After the previous turn has fully completed, sending a new image-only message (or an image + text message) causes the next assistant response to be labeled Steered conversation.

This appears to be a fresh turn, not a same-turn steer.

I can reproduce this with a very small example:

• first send a normal text message like hi
• wait for the assistant to finish
• then send an image-only message so the user bubble shows (No content)

The next assistant response is still labeled Steered conversation.

What steps can reproduce the bug?

1. Start a new chat in the Codex App.
2. Send a plain text message such as hi.
3. Wait until the assistant finishes and the turn is clearly complete.
4. Send an image-only message, or send a mixed image + text message.
5. Observe that the next assistant response is labeled Steered conversation.

What is the expected behavior?

A brand new turn after the previous turn has completed should not be labeled Steered conversation.

Additional information

<img width="820" height="412" alt="Image" src="https://github.com/user-attachments/assets/9d6006d3-09d3-486f-bb8d-c5a33f5cf790" />

This looks like either:

• stale active-turn / steer state being reused after turn completion, or
• a UI labeling bug where multimodal follow-up turns are incorrectly marked as steered