In plain terms
Someone pastes a customer contract into a personal chatbot account to get a summary. Someone else installs a browser extension that reads every page they open. Nobody meant harm and the work got done faster. But company data has left the building, and nobody responsible for it knows.
Why it matters
It is already happening in your organisation; the only open question is how much. The risks are data exposure, regulatory breaches and decisions based on unchecked output. The opportunity is equally real: shadow AI shows you exactly where employees find AI useful, which is the best input an AI roadmap can get.
Example
Picture a mid-sized firm whose official AI assistant has 200 users, while its network logs show more than 600 people reaching consumer AI services every week, many through personal accounts with no data-protection terms.
Most often confused with
Shadow AI vs. Shadow IT
Shadow AI is a branch of shadow IT with two added risks. The tool does not only store data; it reads and may learn from whatever is typed into it. And its output flows back into company work as text, code and decisions that nobody reviewed.
Under the hood
Forms: personal accounts on consumer chatbots, browser extensions and meeting note-takers, AI features switched on inside existing SaaS products, API keys bought on a personal card, and agents or MCP servers that employees run locally with access to company systems. Discovery combines network and SSO logs, expense data and plain surveys. Responses that work combine an approved tool that is at least as good as what people use on their own, a short clear policy on what data may go where, technical controls (DLP, blocking of unmanaged accounts) and a fast route to request new tools.