Shadow IT is bad enough. Add generative AI, and you’ve got a recipe for serious data exfiltration. I’ve seen organizations completely miss this control.
Any licensed user can spin up a Copilot Studio agent that connects to your SharePoint, Dataverse, SQL, or even external APIs. They can publish it to Teams, and suddenly, an innocent-looking HR bot is silently funneling sensitive data out via an HTTP connector. This is a critical governance gap.
The fix is straightforward: disable generative AI agent publishing tenant-wide in the Power Platform admin center. Then, configure environment routing to sandbox environments and use Power Platform environment-level security roles to explicitly control who can create and publish. This isn’t optional; it’s foundational.
To disable tenant-wide publishing, navigate to Power Platform admin center > Environments > Settings > Generative AI > Agent publishing. Turn it OFF for unauthorized makers.
Don’t let your “helpful” internal bots become your next data breach vector. Lock this down now.
The fix
# Check environment creation restrictions
Add-PowerAppsAccount
Get-TenantSettings | Select-Object -ExpandProperty powerPlatform |
Select-Object -ExpandProperty governance
Reference: https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance