ChatGPT Work’s cloud browser stopped working for me on October 29, 2026, and continues to fail on October 30. Browser automation previously worked. It now fails even when asked to open a harmless public page such as example(dot)com.
I’ve reproduced the failure across different browsers, networks, and physical computers, as well as in new Work conversations.
Product and environment
-
Product: ChatGPT Work through the web interface
-
Subscription: ChatGPT Plus
-
Model shown during troubleshooting: GPT-6.1 Sol Light
-
Captured reproduction device: Windows 11 Home, version 25H2, OS Build 26200.9711
-
Browser for the captured reproduction: Google Chrome
-
Exact Chrome version: not recorded
Steps to reproduce
-
Open a new Work conversation.
-
Ask: “Open example(dot)com in the cloud browser.”
-
Work attempts to initialize browser control.
-
The tool crashes before creating a tab or returning browser state.
The same failure also occurs when Work calls await cua.getState();.
Expected behavior
The cloud browser initializes and opens example(dot)com.
Actual behavior
The browser does not open. Attempts return either:
trusted Node process exited unexpectedly; kernel reset, rerun your request
or:
node_repl kernel exited unexpectedly
node_repl diagnostics: {"kernel_pid":3184,"kernel_status":"exited(code=1)","kernel_stderr_tail":"bwrap: Can't create file at /home/agent/.codex/sessions: Read-only file system","reason":"stdout_eof","stream_error":null}
The diagnostics above are from an earlier attempt in the same conversation. Repeated attempts return the same filesystem error with different process IDs.
Timestamp
HAR-captured reproduction on October 30, 2026:
-
Request submitted: 7:07:56.937 a.m. PDT
-
Failure response recorded: 7:08:08.349 a.m. PDT
That captured attempt returned the “trusted Node process exited unexpectedly” message.
Troubleshooting already tried
-
Restarting Chrome
-
Starting new Work conversations
-
Switching browsers
-
Switching networks
-
Switching physical computers
-
Retrying the minimal example(dot)com request
These steps have not restored browser access. Ordinary code/shell commands still work in the affected Work conversation.
I collected a HAR and browser console log and can provide them privately to Support. The most specific diagnostic is the read-only filesystem error at /home/agent/.codex/sessions.
Has anyone encountered this particular cloud-browser startup failure or found a supported recovery procedure?