
Shadow AI is your best requirements document
Two-thirds of professionals use AI tools they think are banned. Before you write the crackdown memo, read what the smuggling is telling you.
Here is the uncomfortable arithmetic of 2026. Sixty-six percent of office professionals have used AI tools at work despite believing they were not permitted. More than a third have pasted customer data into public models. Verizon's DBIR found shadow AI detections up fourfold in a year. And my favorite detail, from a study the executives should read twice: decision-makers use unsanctioned AI at twice the rate of their staff. The people writing the policy are its most enthusiastic violators. I have sat in that steering committee. I know which window was open on the chairman's laptop.
the crackdown memo fails, every time
Banning tools people find useful does not stop the use; it stops the visibility. The PagerDuty survey's grimmest line is that nearly half of workers would rather use AI secretly than risk being told no. A ban converts your adoption problem into a monitoring problem, and monitoring problems are more expensive.
read the smuggling instead
Every unsanctioned tool is a signed requirement. Someone risked a policy violation because the sanctioned way of doing their job was slower. So inventory the shadow usage (amnesty first, or the inventory will be fiction), rank it by data sensitivity, and then do the only thing that works: provide a sanctioned path that is at least as good for the top three use cases, and make the genuinely dangerous patterns technically hard instead of theoretically forbidden.
The inventory conversation itself needs one design choice made in advance: who hears the answers. Run it through line managers and you will collect what people think their manager can bear to hear. Run it anonymously through a survey and you get volume without detail. The format that works is a borrowed neutral: someone outside the reporting line, an hour per team, notes without names attached to tools. People tell the truth about their workflow to anyone who cannot appear in their performance review, and the resulting list is the most honest requirements document your AI program will ever receive.
The companies handling this well treat shadow AI the way sensible IT departments eventually treated shadow spreadsheets: not as a crime wave, but as a backlog someone else prioritized for free.