Survey every department.
Ask which AI tools each one uses, and expect some IT never reviewed.
Ask each department to start the list, then let real use keep it current.
It can stay out of sight until the work gives it away. In late 2023, a lawyer at an Australian state child protection agency noticed unusual wording in a report already filed with a children’s court, and the report turned out to have been drafted with a public AI chatbot. The department’s review then found that nearly 900 of its staff, almost 13 percent, had accessed a public AI chatbot in six months (the state privacy regulator’s report).
Shadow AI is that use, AI for work that nobody reviewed or approved. It arrives two ways. Staff sign up for free accounts. Vendors switch AI on inside software you already bought, and you can’t always turn it off.
Ask, and make it safe to answer, because blocking first sends the use where you can’t see it.
Ask which AI tools each one uses, and expect some IT never reviewed.
A survey that punishes gets a short list.
Free accounts used for work never show up in what you bought.
Have each vendor say in the contract whether its software uses AI.
A survey goes stale as soon as it’s done. Real use keeps it current. Each captured request goes on your record with the person, the device and the AI tool. That helps you meet GOVERN 1.6 in NIST’s AI RMF 1.0, which asks for an inventory of the AI systems you use.

Without VerillianNo policy, no record.
With VerillianYour policy, on the record.
Notice it, find out where the data goes, then decide on it and write the decision down.
See an AI inventory built from real use.