[Windows] ChatGPT and Codex windows can show the wrong conversation/context in multi-window use

Platform

Windows desktop app, current unified ChatGPT/Codex experience.

Issue

When using two desktop windows side by side — one for ordinary ChatGPT and one for Codex — the window/session association can become confused after extended use. A window that visually appears to be the ChatGPT window (including the ChatGPT title/header) can end up displaying the Codex conversation/session, or otherwise show the conversation expected in the other window.

Because the native window still appears to be the expected ChatGPT window, this can go unnoticed. Prompts are then submitted into the wrong conversation/context. In a development workflow this causes Codex to answer questions that belong to a different chat/project context, producing misleading or nonsensical responses and wasting substantial model usage/tokens.

This has occurred more than once, typically after many hours of working with both windows open.

Typical workflow / reproduction attempt

1. Open the Windows desktop app.
2. Keep two windows open simultaneously for an extended session.
3. Use one as ordinary ChatGPT and the other as Codex for a software project.
4. Switch repeatedly between the two throughout the day.
5. Eventually observe that a window can display the conversation/session associated with the other context while its surrounding window identity still suggests ChatGPT/Codex as before.
6. Continue typing without noticing the swap; the prompt is handled in the wrong thread/context.

The issue is intermittent, so exact deterministic reproduction steps are not yet known.

Expected behavior

Each native window must remain bound to the conversation/session and mode selected in that window. A ChatGPT window must never silently surface a Codex thread (or vice versa) because of state/session synchronization between windows. If a thread is intentionally moved or opened in another window, the window identity and visible context should update atomically and unambiguously.

Impact

High for multi-window professional workflows: wrong-context prompts can lead to incorrect project actions/answers and significant unnecessary token/model usage before the user notices the window/session swap.

Related observations

There are already open Windows desktop issues involving multi-window/session identity and the unified ChatGPT/Codex desktop app, including generic/misleading native window titles and thread-switching/resume regressions. This report is specifically about apparent cross-window conversation/context switching, which is more severe than identical window titles alone.

No private conversation contents, repository data, account identifiers, or local paths are included in this report.

Start with the Windows desktop app's multi-window workflow: keep ChatGPT and Codex windows open, switch between them repeatedly, and investigate when their visible conversation or mode diverges. Compare the window identity with the displayed session after switching or resuming. Done means each native window stays bound to its selected conversation and mode, or updates both atomically when intentionally moved.