Shadow AI: Find It Before It Finds Your Data
Shadow AI is any AI tool used at work without approval. Here is how to find it, set an allowed-tool list and agree data rules without slowing your teams down.
In this article
Shadow AI is the use of AI tools at work without the knowledge or approval of the people responsible for security. A marketing manager signs up for a free writing assistant. A developer installs a browser extension that summarises code. A support agent pastes a customer complaint into a chatbot to draft a reply.
None of these people are trying to cause harm. They are trying to work faster. The risk is that company and customer data ends up in tools nobody has checked, under terms nobody has read. This guide explains how to find shadow AI, how to set an allowed-tool list and how to agree data rules that people will respect.
1. Why shadow AI is a real risk
The main risk is data. When someone pastes information into an AI tool, it leaves your control. Depending on the tool and its settings, that data may be stored, used to improve the service or reviewed by the provider’s staff. If it includes personal data, confidential plans or source code, you may have a problem under your contracts, your security policies and data protection laws such as India’s DPDP Act.
There are other risks too. AI output can be wrong, and staff may rely on it without checking. Some tools ask for broad permissions to read email or files. Browser extensions can see everything on a page. And because nobody approved the tools, nobody is watching for changes in their terms or security.
The aim is not to stop people using AI. It is to bring that use into the open, where it can be managed.
2. How to find the AI already in use
You will not find every tool, and you do not need to. The goal is a realistic picture of how AI is used, so you can make sensible decisions.
Ask people directly.
A short, friendly survey asking which AI tools people use and why often reveals more than any technical scan. Make it clear the aim is to help, not to punish.
Check sign-ins and expenses.
Look at single sign-on logs, app integrations and expense claims for AI subscriptions. Paid tools often appear on corporate cards first.
Review browser extensions.
Many AI tools arrive as extensions. If you manage company browsers, list the extensions in use.
Look at network and endpoint data.
Where you already have web filtering or endpoint tools, review traffic to known AI services, in line with your privacy notices to staff.
Talk to team leads.
Ask managers which tools their teams rely on. They often know which workflows depend on AI.
Record what you find in a simple register: the tool, the team, the purpose, the data involved and a first view of the risk. This register becomes the starting point for your allowed-tool list.
3. An allowed-tool list and data rules
Once you know what is in use, sort the tools into three groups: approved, approved with conditions, and not approved. Approval should depend on the vendor’s data handling terms, the security settings available, where data is stored and whether an enterprise plan with stronger protections exists.
Publish the list where people can find it, with a short reason for each decision. When you reject a popular tool, suggest an approved alternative. People accept a "no" more easily when there is a "use this instead".
Pair the list with a few clear data rules. A simple version might say: public information can go into any approved tool; internal information only into tools approved for internal data; confidential and personal data only into tools approved for that purpose, with enterprise settings turned on. Use the data categories your organisation already has, so people are not learning a new scheme.
Make the request route fast. If someone needs a new tool, a short form and a quick review keep them on the approved path. If approval takes weeks, shadow AI will return.
Review the register and the list regularly. AI tools change their features and terms often, and new tools appear all the time. A short monthly check by the owner keeps the list honest.
Finally, connect this work to your wider AI governance. Our article on a practical AI governance framework explains how inventories, policies and reviews fit together. If you are working towards ISO 42001, the shadow AI register becomes part of your AI system inventory.
If you would like help discovering the AI already in use and setting sensible rules, our ISO 42001 consulting team is happy to help.