Picture a ChatGPT Enterprise workspace a year after rollout. The IT lead who bought the licenses is still an Owner. So are the two admins she added during a busy week, and the director who asked for access “just to check something.” A contractor who left in the spring still has an account, because he was added by hand instead of through the identity provider. And a service account someone set up to run Codex jobs has a credential that never expires.
None of this is unusual. It’s what happens when a tool goes from pilot to every department in a few months, and nobody’s job is to keep an eye on the admin console.
Most of the security conversation around ChatGPT is about what people paste into it, and that’s fair. OpenAI’s Compliance Platform lets DLP and data security tools read conversations. But a more basic question usually goes unanswered: who actually holds the keys to the workspace?
Why nobody owns this
ChatGPT Enterprise falls between teams, so each one assumes someone else is watching it:
- Identity teams own the identity provider, but SCIM provisioning to ChatGPT is optional. Anyone added directly in the admin console never shows up in their access reviews.
- Cloud security teams own AWS, Azure, and GCP. A ChatGPT workspace isn’t a cloud account, so their CNAPP never sees it.
- The workspace owner is usually whoever bought the licenses. Owner access gets handed out to get things done, and it rarely gets taken back.
Rolling out an enterprise AI platform is still the right call. Verizon’s 2026 DBIR found shadow AI is now the third most common non-malicious insider action, up 4x in a year, and a sanctioned tool is the best way to get people off personal accounts. But someone still has to check how that tool is set up.
What Orca checks in ChatGPT Enterprise
You connect ChatGPT Enterprise to Orca with one read-only Admin API key. There’s nothing to install and no change to how anyone uses ChatGPT. Orca pulls in every user, group, service account, and service-account credential in the workspace, and flags issues like these out of the box:
| Orca detects | Why it matters |
| A user with the workspace Owner role | Owners control every setting, member and key in the workspace. Each one is a high-value account |
| A workspace with more than three owners | Owner access tends to spread and rarely gets revoked. The same rule applies to Azure subscriptions and Supabase |
| A service-account credential with no expiration date | A token that never expires stays valid long after the job or person that needed it is gone |
| A service-account credential not rotated in 90 days | OpenAI suggests token lifetimes of 7 to 90 days. Older tokens have had more time to leak |
| A user not provisioned through the identity provider | Users added by hand skip your joiner-mover-leaver process, so offboarding may never reach them |
| A group managed by hand instead of synced from the identity provider | Manual groups drift from the access you approved in the IdP |
| A group with no members | Empty groups are leftovers that can quietly grant access to the next person added |
Screenshot caption: A ChatGPT service-account credential with no expiration date, shown in Orca with its workspace, creation date and remediation steps.
ChatGPT belongs next to the rest of your risk
An AI-built app has five parts: the model behind it, the agent or builder that made it, the platform it runs on, the database and login behind it, and the cloud account under it. Each comes from a different vendor and has a different owner. None of those owners is security.
Orca AI AppGen Security brings those parts into one platform. ChatGPT Enterprise sits next to Claude Enterprise at the model layer, alongside Vercel and your AWS, Azure, and GCP accounts.
That matters because OpenAI shows up in a lot more places than the ChatGPT admin console. At most companies, it looks something like this:
- ChatGPT owners get reviewed in a spreadsheet once a quarter, if at all.
- A DLP tool reads chats, a secrets scanner finds OpenAI API keys in repos, and the CNAPP covers Azure OpenAI.
Each tool sees its own piece, so nobody can say how exposed the company is to OpenAI overall. In Orca, ChatGPT users, groups, and credentials sit in the same inventory as OpenAI API keys found in code and container images, workloads calling OpenAI models (seen by the Orca Sensor), and Azure OpenAI resources. They land in the same queue as your Claude Enterprise and Vercel findings, too.
Screenshot caption: The Orca AI Security dashboard showing ChatGPT Enterprise and Claude Enterprise workspaces side by side with cloud AI services.
The goal isn’t to slow anyone down. Your builders keep using ChatGPT the way they already do, and security finally gets a clear look at who holds the keys, without reading a single prompt.
Want to see your ChatGPT Enterprise workspace in Orca? Book a demo today.
About the Orca Platform
Orca Security delivers security for the companies that build. As cloud, applications, AI and app generation expand the attack surface, Orca transforms security risk into the context teams need to act. The Orca Platform provides complete visibility across cloud, AI, and application environments, correlates risk across every layer, and prioritizes the exposures that matter most so organizations can remediate faster and innovate with confidence. Trusted by hundreds of organizations, including SAP, Autodesk, Gannett, Lemonade, and Digital Turbine, Orca is backed by leading investors including Temasek, CapitalG, ICONIQ Capital, and Redpoint Ventures.
Chat with an Orca Expert