Shadow AI is when staff use AI tools at work without approval or visibility. To find out what AI your staff are using, combine IT discovery with conversations. Logs show which tools get opened; surveys get the safe answer. One-to-one AI interviews, like Ethnobot's, reveal which tools people use, for which tasks, the workarounds, and the risks.
Monitoring can show which sites get visited. Ethnobot is the human half of the picture — one-to-one interviews that surface the tools and workarounds people describe, alongside IT discovery.
Try the interview yourself first. Free, no signup.
Shadow AI is staff using AI tools at work without the organisation's approval or visibility. It shows up in plain ways: drafting proposals in a personal ChatGPT account, pasting data into a free tool, or installing a browser AI plug-in that never went through IT.
Shadow IT is unapproved software or devices. Shadow AI goes further — it processes what people type into it. Bring-your-own-AI (BYOAI) is the same pattern under another name.
People hide AI use when policy is unclear, approved tools are slow or missing, or they fear looking less capable. Tick-box data cannot tell you about the consultant who secretly uses AI to write proposals and feels conflicted about it.
That gap — people can do more with AI than the organisation enables — is the Absorption Gap. Shadow AI is often the Absorption Gap showing up as secret workarounds.
Read the Absorption Gap framework →Risks people describe include data leaving the organisation, unchecked outputs, and uneven practice across teams — useful signals for policy and enablement.
Wins sit in the same interviews: the hidden experts the organisation hasn't noticed, the hesitant people who need clearer permission, and the patterns you didn't know existed. Banning shadow AI loses the wins as well as the risks.
See the six AI archetypes →Logs alone aren't enough. You also need to hear from people, in a way they'll answer openly. A one-to-one AI interview gets the answer a survey cannot.
IT discovery — sign-ins, network logs, expense claims — shows which tools appear. Useful for inventory. It cannot show why someone used a tool, for which task, or what they were afraid to admit.
Surveys get safe answers. “Rate your AI confidence 1–5” gives you a number, not a story. A one-to-one AI interview follows up, gets examples, and records the person's own words — including the tools and workarounds people describe.
Across the team: the tools and workarounds people describe, for which tasks, where people are stuck, which risks recur, who the hidden experts are. That is the signal for policy and enablement — not another seat count.
| Approach | What it shows |
|---|---|
| Monitoring / IT discovery | Which sites or tools get opened |
| Survey | A rating or tick-box — often the safe answer |
| Ethnobot interview | The tools and workarounds people describe, plus tasks and reasons, in the person's own words |
Use discovery and interviews together.
An org study returns cohort-level themes, segment profiles, a quote library (quotes are never attributed without consent), plus a Leadership Brief and Team Debrief. These are live deliverables on /org-study.
Sample report figures are an example — not a named client case.
Staff using AI tools at work without the organisation's approval or visibility.
It can expose data and creates uneven practice, but it also hides the people who are already getting value from AI.
Combine IT discovery with one-to-one conversations; logs show tools, interviews show tasks, reasons and risks.
Surveys tend to get the safe answer; people under-report use they think their manager would disapprove of.
Banning tends to push use further underground; find out what's happening first, then set policy and approved tools.
Org Study findings are reported at cohort level, and quotes are never attributed without consent.
Detection shows which tools are used; Ethnobot shows how, why and what people need. Use both.
Start with your own profile, pulse one team, or commission a full organisational study.
Commission AI Adoption Intelligence →