Windows: Remote Control toggle immediately turns off; repeated HTTP 503 and misleading instance warning

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

Windows app package: 26.903.8094.0 (OpenAI.Codex). Bundled Codex CLI: 0.153.4. Version identified from the running installation, not the About dialog.

What subscription do you have?

ChatGPT Pro

What platform is your computer?

Microsoft Windows NT 10.0.26200.0 x64 (Windows 11)

What issue are you seeing?

Remote Control cannot be enabled on this Windows PC. In Settings > Connections > Control this PC, turning on Allow connections immediately switches it back off. The failure occurs before another device can connect.

The Portuguese UI shows:

• Não foi possível atualizar a disponibilidade do controle remoto.
• On earlier attempts: Não foi possível ativar o controle remoto. Verifique se há apenas uma instância do ChatGPT em execução.

English translations: Could not update Remote Control availability; could not enable Remote Control, check that only one ChatGPT instance is running.

Local process inspection found only one main app instance; the other ChatGPT processes are its renderer/GPU/network children. The app-server is a child of that same instance.

The app-server logs repeatedly show HTTP 503 Service Unavailable when connecting to wss://chatgpt.com/backend-api/wham/remote/control/server, with the response:

upstream connect error or disconnect/reset before headers. reset reason: connection termination

The available log sample contained 389 HTTP 503 failures between 2026-09-09 23:37:57 UTC and 2026-09-10 04:58:14 UTC. This is the observed log interval, not necessarily the incident onset. Enrollment is already present (has_enrollment=true). This suggests a service or connection-route failure; it does not establish a global outage. The duplicate-instance warning appears misleading for the current failure.

What steps can reproduce the bug?

Feedback ID: 01a07a2e-6d73-70b3-a297-9c9a7806e74c

What is the expected behavior?

Allow connections should remain enabled when registration and the remote connection succeed, so another device can connect. If activation fails because the remote service returns HTTP 503, the UI should report the connection/service failure rather than suggesting another app instance without evidence.

Additional information

Read-only diagnostics on 2026-09-10:

• codex doctor run outside the sandbox confirmed configured authentication, healthy local databases, successful desktop app-server initialization, and reachable provider HTTP endpoints.
• The separate Responses WebSocket probe also failed; it was not treated as proof of the specific Remote Control cause.
• No proxy environment variables, no WinHTTP proxy, and the Windows user proxy is disabled.
• The same device has earlier successful Remote Control WebSocket connections in the available logs at 2026-09-09 00:19:42 and 00:30:40 UTC.
• Example failed request for support correlation: 2026-09-10 04:57:50 UTC; cf-ray a38bdaac298c7e15-GRU.
• Another-network comparison could not be performed: the PC currently exposes only Ethernet and USB tethering with the iPhone was not available.

No credentials, full logs, private project content, device enrollment IDs, or screenshots containing personal information are attached. Please investigate the Remote Control endpoint/route and the misleading duplicate-instance message.

Start by tracing the app-server handling for wss://chatgpt.com/backend-api/wham/remote/control/server and the UI path that reports activation failures. Reproduce the HTTP 503 behavior using the supplied feedback ID and logs; done means the toggle and error message correctly distinguish a service or connection failure from a duplicate-instance condition.