Shadow AI: The tools nobody approved
Feb 5th, 2025 by Jesper Kråkhede
A few weeks ago a new AI chatbot from China shot to the top of the app stores. Within days, researchers found one of its databases exposed on the internet, including chat histories and other sensitive information. And in many organisations, employees had already started pasting work documents into it.
This is shadow AI. Just like shadow IT ten years ago, people use the tools that help them do their job, whether IT has approved them or not. The difference is that with AI, the thing they upload is often exactly the sensitive data: contracts, code, customer lists, strategy documents.
What can you do technically? First, find out what is used. Defender for Cloud Apps can discover which AI services are accessed from your network and devices, and how much data goes to them. Many organisations are surprised by the list.
Second, decide what is allowed and enforce it. Sanction the services you trust, block the ones you don´t, and use the endpoint and the browser to stop data being pasted or uploaded where it shouldn´t go. Data loss prevention can recognise sensitive content and stop it before it leaves.
Third, label your data. Sensitivity labels follow the document, so you can protect the confidential material regardless of which tool someone tries to use.
And fourth, give people a good alternative. Blocking without offering a safe tool just makes them find another way around.
You can´t stop people from using AI. You can decide which AI they use.