The Admin plugin for ChatGPT Work: a real governance layer, or just a chat UI over settings?

What it is

A workspace-admin surface that lives inside ChatGPT itself. OpenAI released an Admin plugin for ChatGPT Work and Codex on Tuesday, allowing workspace administrators to analyze information, update settings, and process requests within a single conversation. Instead of clicking through the admin console, an admin asks questions in chat and the plugin returns usage data or performs authorized changes.

The next-work-session test

You're the consultant running a Monday steering committee for a 4,000-seat ChatGPT Work rollout at a regulated client. The CIO wants three things: which BUs are actually using Codex, who's near their credit ceiling, and evidence that offboarded contractors lost access. Today that's three tabs and a data pull. With the plugin, per OpenAI's own demo, in a demo of the new plugin, Aarti Bagul, a Member of the GTM Staff at OpenAI, shows a use case in which she is preparing for a monthly AI rollout review — the same shape of question your steering deck needs. It changes what you scope into the "governance workstream" of a transformation SOW: less bespoke reporting, more policy design.

Pricing

There is no separate price for the plugin itself. It ships as an admin capability inside ChatGPT Work and Codex, which are sold under OpenAI's business plans. OpenAI publishes business plan pricing and separate credit rate cards for Business and Enterprise/Edu, and credit-based pricing and token-based pricing are available for Enterprise plans. Enterprise itself is quoted per organization — ChatGPT Enterprise is purchased at the organization level. Individual users do not sign up for Enterprise directly. Public list pricing for Enterprise seats: unverified. Confirm with the OpenAI sales team for any client proposal.

What we'd actually use it for

Three concrete things in a deployment engagement:

Limits

Skepticism warranted on a few points.

• It's not a replacement for your GRC stack. Audit-event coverage is still governed by workspace role and scope: workspace owners and admins can now review supported audit events from the Global Admin Console. Access and event coverage depend on the selected workspace and admin role. "Supported audit events" is doing work in that sentence — expect gaps against SOX or FedRAMP evidence requirements.
• It inherits, not overrides, existing policy. Behind each request, the plugin ... to ChatGPT Work and Codex as permission-aware tools, while honoring workspace policies and approval requirements. Good for safety, but it means a badly configured workspace will produce badly scoped answers. Your policy design work doesn't go away.
• Natural-language admin is faster, not safer. A chat prompt that reads "remove all contractors from Group X" is one typo from an incident. Expect clients to demand a second-approver flow before enabling destructive actions.
• Codex-specific controls still evolving. Per OpenAI, Codex seats are governed by workspace-level admin controls, just like standard seats — fine in principle, but Codex usage patterns (long-running agents, repo access) don't map cleanly to seat-based analytics yet.
• No pricing transparency for procurement. You will still be negotiating Enterprise commercials the old way.

Try it if

• You're advising on a ChatGPT Work or Codex rollout above ~500 seats and need repeatable adoption reporting.
• Your client's admin team is small and drowning in console clicks.
• You're building a monthly AI governance ritual (steering deck, credit review, access review) and want the data pull automated.

Skip it if

• Your client is on ChatGPT Plus, Team, or a non-Work plan — the plugin doesn't apply.
• Governance is owned by a separate GRC or IAM platform your client has already standardized on. This won't replace it.
• You need immutable, exportable audit trails for a regulator today. Validate coverage against the admin controls documentation before you make that promise in a SOW.