OpenAI starts Enterprise preview of ChatGPT Sites controls and private-network access

OpenAI began rolling out an admin preview of ChatGPT Sites to eligible Enterprise workspaces on October 15, 2026. The update pairs controls over who can create and share Sites with private-network connections and regional hosting. Administrators can review access before enabling Sites for members, while public publishing remains a separate, off-by-default choice.

Private connections do not grant publishing rights

Tunnels, which are also rolling out to eligible Enterprise workspaces, connect Sites to services on a company’s private network. Those services do not have to become public for a Site to reach them.

Permission to connect a tunnel does not also grant permission to publish a tunnel-using Site publicly.

Hosting follows the region only in the US and EU

For workspaces with US or EU data residency—a setting governing data location—new Sites automatically use the workspace’s existing region for supported storage and hosting.

Existing Sites do not automatically move. OpenAI also distinguishes Sites regional hosting from inference residency, which concerns where model processing happens.

Connected apps retain a separate permission boundary

OpenAI described this connected-app capability in a September 29 update. Enterprise members could build private Sites with supported apps, including issue dashboards. Each member’s own connected account and permissions apply, and the member approves read-only access.

• Organization approval is required for connected-app use in Sites. Workspace admins must also enable Allow use in Sites for each plugin; that setting is off by default.
• A visitor’s connected-account access is available while they use the Site. Background scheduled tasks cannot reuse that visitor access, keeping interactive viewing distinct from unattended work.