Ask a leadership team whether their company uses AI, and you'll usually get a careful answer about the pilot in progress, the vendor under evaluation, the committee reviewing use cases. Ask the people actually doing the work, and the answer is different: most of them are already pasting documents into a chatbot, using it to draft emails, or running data through a tool nobody in IT has heard of. The gap between those two answers is shadow AI, and it's larger in most organisations than anyone at the top expects.
The AI policy usually arrives after the AI usage, not before it.
How it actually shows up
- Personal accounts on free tools, used for work tasks because the approved option is slower or doesn't exist yet.
- Browser extensions and plugins that quietly send page content or documents to a third-party model.
- One team's "unofficial" workflow, built by someone who got tired of waiting for a sanctioned tool and just solved their own problem.
- Client or company data pasted into a prompt because it was the fastest way to get an answer, with nobody weighing what that data actually was.
Why banning it doesn't work
The instinct after discovering shadow AI is usually to lock it down: block the domains, restrict the extensions, send out a policy memo. That solves the visible half of the problem and does nothing about the half that's already invisible. People adopted these tools because they were solving a real problem faster than the approved path did. Remove the tool without replacing the reason it existed, and the workaround doesn't disappear, it just gets harder to see.
What a workable response looks like
- Find out what's actually being used before writing any policy — a short, no-blame survey usually surfaces more than IT logs do.
- Sort what you find by data sensitivity, not by tool popularity: a free tool used on public information is a very different risk than one used on client data.
- Give people a sanctioned path that's at least as fast as the workaround, or the policy will just get ignored quietly rather than followed.
- Treat the discovery as ongoing, not a one-time audit — new tools show up faster than any policy review cycle.
The organisations that get ahead of this aren't the ones with the strictest policy. They're the ones who found out what was actually happening early enough to shape it, instead of discovering it during an incident review. Shadow AI isn't a sign that people are careless. It's a sign that the sanctioned path is slower than the problem it's supposed to solve, and that's a gap worth closing before someone else closes it for you.