ChatGPT Work DNS fails in Windows and Linux executors, but works outside ChatGPT. Anyone else experiencing this?

English isn’t my native language, so I asked ChatGPT to help me write this post. The test results and errors are from my actual troubleshooting. I’d really appreciate any help or suggestions. Thank you!

I’m using ChatGPT Plus in my personal workspace and troubleshooting a network issue that prevents me from publishing a private application through Sites.

What happens

DNS resolution fails inside ChatGPT Work for both:

• git.chatgpt-team.site
• registry.npmjs.org

However, both domains resolve successfully when I run Resolve-DnsName in PowerShell opened directly on my computer, outside ChatGPT.

Where I reproduced it

• Windows executor in the original Work conversation.
• Windows executor in a separate, new Work conversation.
• Linux container in a new Work conversation through the web browser.

Errors

Windows:

ENOTFOUND 

or:

No such host is known 

Linux:

[Errno -3] Temporary failure in name resolution 

The tests fail before HTTPS, so I have no HTTP response to report.

What I have checked

• “Work network access” is enabled in Settings → Data controls.
• Changed the conversation’s approval mode to “Ask for approval.”
• Closed and reopened the desktop app.
• Reproduced the issue in new conversations and in the browser.
• Tested with Node.js, curl and .NET in the Windows executor.
• Submitted feedback and requested escalation to a support specialist.

In the Windows session, an explicit request for elevated execution was rejected with a policy reporting:

sandbox_approval: false request_permissions: false 

I understand that this may be separate from the Linux container’s network permissions. I don’t have an organizational Admin Console because this is a personal Plus workspace.

My questions

1. Has anyone experienced this with Work network access already enabled?
2. Is there a supported way to inspect the effective network policy for a personal Plus account?
3. Is there a documented way to refresh the execution environment so it picks up the network setting?

I haven’t established whether this is an expected restriction, an account configuration issue or a bug. Any comparable experiences or documented troubleshooting steps would help.

Tests performed on October 22, 2026.