Shadow AI: why employees use unapproved AI tools
Shadow AI is staff using AI tools you haven't approved, like personal ChatGPT accounts. Why it happens, why bans fail, and how to find out without blame.
Shadow AI is employees using AI tools their organisation hasn’t approved, such as a personal ChatGPT account, for work. It usually happens because the approved tool is missing, slower or worse at the task, or because people don’t know what is allowed. A ban on its own pushes the use out of sight. The better response is to find out what people use and why, without punishing them, and then close the gap.
What is shadow AI?
The term borrows from “shadow IT”, the spreadsheets and cloud apps that teams adopt without asking IT. Shadow AI is the same habit with AI tools. Some examples from ordinary work:
- A sales manager drafts proposals in a personal chatbot account because the company tool doesn’t know the product range.
- A care coordinator pastes a weekly rota into a free tool to rebalance shifts.
- An analyst installs a browser extension that summarises every page they open, including internal ones.
- A team lead uses an app on their own phone to transcribe one-to-ones.
None of these people is trying to cause harm. They are trying to get work done. The risks come from what the tools do with the data and from the fact that nobody else knows: customer or patient information leaves your systems, there is no record of what was produced, and wrong answers go out unchecked.
How common is shadow AI?
Very. In the 2024 Microsoft and LinkedIn Work Trend Index, 75% of knowledge workers said they use AI at work, and 78% of those AI users said they bring their own AI tools to work. BCG’s AI at Work 2025 survey found that when employees lack the AI tools they need, more than half say they will find alternatives and use them anyway.
So if your Copilot or company chatbot figures look low, some of that work is probably happening elsewhere. Our guide to measuring Copilot adoption beyond licence data explains why usage reports can’t see it.
Why do employees use unapproved AI tools?
When you ask, the reasons tend to be practical.
There is no approved tool, or it arrived late
People started using public chatbots at home and brought the habit to work long before most companies picked a tool. Some roles still have nothing approved.
The approved tool is worse for their task
It may be slower, limited to certain apps, or weaker at the thing they need, such as translation, coding or working with long documents. If the personal tool gives a better answer in half the time, people use it.
Nobody told them clearly what is allowed
Gallup’s 2025 research found only 30% of employees said their organisation had general guidelines or formal policies for using AI at work. Without a clear rule, people decide for themselves.
They don’t want to be seen using it
The same Work Trend Index found 52% of people who use AI at work were reluctant to admit using it for their most important tasks. A personal account on a personal phone is the easy way to keep it quiet.
Why don’t bans work on their own?
A ban removes the tool from the company laptop. The need stays. Someone who saved an hour a day with a chatbot will often move the work to their phone, where you have even less sight of what data goes in. The use doesn’t stop. It just stops showing up.
Bans also teach people that talking about AI is risky. After that, they won’t tell you what they use, and you lose the information you need to choose better tools and write sensible rules. Some blocks make sense, for instance on tools that keep and train on what is entered. They work best paired with an approved option that does the job and a plain policy people can follow.
For a structure, many organisations start from the NIST AI Risk Management Framework, a voluntary framework released in January 2023, with a Generative AI Profile added in July 2024 to help organisations identify risks specific to generative AI. A framework only helps if you know which tools are actually in use, and for that you have to ask.
How do you find out what people use, without punishing them?
Network and endpoint logs can show which AI websites are visited from company devices. They can’t show personal phones, what people put in, or why they chose that tool. For that you need to ask, and people will only tell you if it is safe to.
- Say up front that the point is to learn what people need, and that nobody will be disciplined for what they report.
- Report only at group level, with a minimum group size. Our post on the employee survey anonymity threshold covers how to set it.
- Ask about tasks first and tools second: “What did you last use AI for at work?” then “Which tool did you use, and why that one?”
- Ask what the approved tool would need to do for them to switch.
- Share what you learned and what you changed, so the next round gets honest answers too.
Whether people answer honestly depends on how safe it feels to admit a mistake at work. If you are unsure, our post on how to measure psychological safety is a good place to start. For a fuller question list, see AI adoption survey questions.
Spradley runs these as one-on-one AI interviews, by text or voice, that people answer on their own phone or computer in their own time. The interviewer asks a follow-up on each answer, which is how “I use something else sometimes” becomes the tool and the reason. Answers are grouped into themes with quotes, departments can be compared, and numbers and ratings are never shown for fewer than five people. See our AI adoption page.
What should you do with what you learn?
Treat shadow AI as demand you didn’t know about. If a whole department translates documents with a personal tool, that tells you what the approved tool must do. If people paste client data in because nobody said not to, the policy needs a clear line on data. If the reason is that people feel watched, look at how managers talk about AI, and at why employees avoid the approved tools in the first place.
Frequently asked questions
What is shadow AI?
Shadow AI is employees using AI tools for work that their organisation hasn't approved, such as personal ChatGPT accounts, free browser extensions or apps on their own phones.
Is shadow AI a security risk?
It can be. Company or customer data may leave your systems, there is no record of what was produced, and wrong answers can go out unchecked. The bigger problem is often that leaders can't see it.
Should we ban ChatGPT at work?
Blocking specific risky tools can make sense, but a ban on its own tends to move the use to personal devices. Pair any block with an approved tool that does the job and a clear policy.
How do we find out which AI tools employees use?
Logs show some of it. To learn the rest, and why, ask people about the tasks they use AI for, promise no blame, and report only for groups large enough that no one can be identified.
Key terms
- shadow AI
- Employees using AI tools for work that their organisation has not approved, such as a personal chatbot account, a browser extension or an app on their own phone.
- shadow IT
- Spreadsheets, cloud apps and other software that teams adopt for work without asking IT. Shadow AI is the same habit with AI tools.
- NIST AI Risk Management Framework
- A voluntary framework from the US National Institute of Standards and Technology, released in January 2023, for managing AI risks. A Generative AI Profile was added in July 2024.
- psychological safety
- A team’s shared belief that it is safe to take interpersonal risks, such as admitting a mistake, asking for help or disagreeing. Amy Edmondson defined and measured it in 1999.
- minimum group size
- The smallest number of responses a group needs before its results are shown. Five is a common default; smaller groups are hidden or combined with a larger one.
