AI permissions are the new shadow IT

Article link: https://open.substack.com/pub/johanos...

The article argues that AI has revived shadow IT in a more dangerous form. In the past, shadow IT was mainly about unapproved apps and scattered data. Now, AI tools are being connected directly to core systems like email, calendars, cloud drives, CRMs, and sometimes finance platforms, often through a single casual “Connect your account” click. Because these tools can read, summarise, and increasingly take actions on a user’s behalf, the risk isn’t just where data is stored — it’s who (or what) has access, and how widely that access has been granted across an organisation.

It then explains why this problem is growing: permission sprawl, agentic AI, and poor organisational visibility. Many users don’t understand what they’ve authorised (read vs write access, folder vs entire drive, personal vs corporate scope), and many organisations don’t have a clear inventory of which AI tools are integrated where. The takeaway is practical: treat AI adoption as an access and identity issue, build visibility into integrations and permission scopes, enforce least privilege by default, and make approvals easy so staff don’t bypass governance. Otherwise, the next major incidents may come not from sophisticated hacking, but from well-meaning people accidentally granting “master keys” to the wrong tool.