Guide · Your team and AI
Shadow AI: why staff use unapproved AI tools, and what to do about it
Shadow AI is staff using AI tools their employer hasn’t approved, often on personal accounts. Most people do it because the tool helps and nobody has said what’s allowed or offered an approved option. The fix is to provide one: an approved tool, plain rules on what data goes where, and a no-blame week to learn what’s in use.
By Andrew Ward, Founder of ScorchsoftChecked 11 minute read
In short
- Treat shadow AI as a sign that people want help with their work and haven’t been given a safe way to get it, not as misconduct.
- Ask before you rule: run a no-blame “tell us what you’re using” week, so your rules fit the work people really do.
- Give people a better option than their own account: one approved tool on an organisation plan, and clear rules on what data may go into it.
- Reach for monitoring last, if at all. The ICO expects any monitoring of workers to be necessary, proportionate and explained to them first.
Written for: Managers, owners and HR or data protection leads in UK organisations of about 10 to 250 people.
What is shadow AI?
Shadow AI is the use of AI tools at work that the organisation hasn’t approved, or doesn’t know about. It usually means a free or personal account on a public chat assistant, used for real work on a phone or a home laptop.
The term borrows from “shadow IT”, the older name for software staff adopt without the IT team. You may also see “bring your own AI” (BYOAI). In this guide, shadow AI means everyday unapproved use by people trying to do their jobs, not deliberate misuse.
The risk sits in the account, not the person. A personal AI account is governed by the terms its holder agreed to, not by your organisation’s. Depending on the tool and its settings, what goes in may be kept, reviewed or used for training, and it leaves with that person when they leave. An organisation plan puts the same tool under terms your organisation chose.
No policy is still a policy
If you haven’t approved a tool or written a rule, your policy in practice is whatever each person decides on the day. That is the line Module 7 of The AI Stairway™ opens its leaders’ section with.
Why do staff use unapproved AI tools?
Mostly for three ordinary reasons: the tool saves them time, nobody has told them what’s allowed, and there is no approved option to use instead. Hidden use is a reasonable response to unclear rules, so treat it as information about what your team needs.
- It helps. A first draft of a tricky email, a summary of a long document, a formula that finally works. People keep using what saves them time.
- Nobody said what’s allowed. In Deloitte’s UK survey, 65% of people who use generative AI reported a lack of convincing leadership on how it should be used in their organisation (25,000 UK workers aged 18 to 70, fieldwork May to June 2026; consultancy research).
- There is no approved option. In Microsoft’s UK survey, one reason employees gave for using unapproved AI tools, cited by 28%, was that their company doesn’t provide a work-approved option (2,003 UK employees, October 2025; vendor research, run by Censuswide).
- It feels safer to say nothing. In the same Deloitte survey, nearly a quarter (23%) of workers surveyed thought there was a stigma attached to using generative AI at work. If people expect to be judged, they keep quiet.
Hayley coordinates client services at a 40-person accountancy practice. She has used AI every day for months to draft chasers for missing paperwork, on her personal account, and has never mentioned it. The chasers are excellent. Then she goes on leave for two weeks and the practice slides back to the old way, because nobody knew where her prompts lived. Nobody had told her what was allowed, and the practice had no approved tool, so she did what seemed sensible. Her manager’s problem isn’t Hayley. It’s the gap she filled.
How common is shadow AI in the UK?
Common, but not universal. Surveys from 2025 and 2026 suggest that most UK employees who use AI at work have used an unapproved tool at some point, while about half of all employees still don’t use AI in their job at all.
| What was measured | Figure | Who was asked, and when | Source |
|---|---|---|---|
| Have ever used unapproved consumer AI tools at work | 71% (51% still do every week) | 2,003 UK employees, October 2025 | Microsoft UK and Censuswide (vendor research) |
| Use generative AI without their employer’s knowledge | 31% of users | Generative AI users within 25,000 UK workers, May to June 2026 | Deloitte UK, fieldwork by Ipsos (consultancy research) |
| Never use AI as part of their job | 52% | 5,074 UK working adults, December 2025 to February 2026 | CIPD Good Work Index 2026, with YouGov |
Read the first figure with the last. “Ever used” isn’t the same as daily use, and the Microsoft survey was published by a company that sells an approved alternative. The honest summary is that hidden use is widespread among people who use AI, and many of your colleagues may not use it at all. Both groups need the same thing from you: a clear, safe way to start or carry on.
What are the real risks of shadow AI?
The main risks are data going where you didn’t choose, work you can’t check or repeat, and know-how locked in one person’s account. None needs alarm, and each has a plain fix.
| The risk | Why it happens | The fix |
|---|---|---|
| Client or personal data goes into a tool under terms you never agreed | Personal accounts follow the individual’s terms and settings | An approved tool on an organisation plan, and tramlines for what data may go in |
| Nobody can say what has gone where | Use is hidden, so there is no record | A no-blame week to find out, then an approved route that people prefer |
| Quality nobody checked | Drafts go out without anyone knowing AI wrote them | A team habit: say what AI did and what you checked |
| Know-how leaves with the person | The best prompts live in one personal chat history | Write the best ones down once and share them as an AI skill, with an owner |
The data risk is the one to take most seriously. Under UK GDPR, a “personal data breach” includes unauthorised disclosure of personal data, so client or staff details pasted into an unapproved tool may need assessing as one (see below). That is a reason to give people a safe route, not a reason to frighten them.
What should you do about shadow AI this month?
Find out what’s in use without blame, approve a tool that meets the same needs, and agree in writing what data may go where. Most small organisations can make all three moves in a month without a budget beyond a few licences.
Step 1: Run a “tell us what you’re using” week (owner: the manager or owner)
Ask everyone, anonymously if that helps, which AI tools they use, for which tasks, what would help and what worries them. Promise in writing that nobody will be disciplined for what they report, and keep that promise.
Step 2: Approve one tool, on an organisation plan (owner: whoever holds IT and data protection)
Pick the tool that meets the needs the week revealed. Before you approve it, run the trust check from Module 6: who runs it, where your data is stored, whether it trains on your data, how long it’s kept, which terms apply, what it can reach and do, and whether your data protection lead would approve this use for this data.
Step 3: Agree the tramlines and a one-page team rule (owner: the team, with its manager)
The tramlines, taught in Module 1, are your rules for what goes into AI: green (go, with nothing private in it), amber (with care, or with approval) and red (don’t). A sensible default is to treat client, personal and special category data as red until your data protection lead says otherwise. Write them on one page with the approved tools, who owns shared prompts and how to report a problem, and set a review date.
What to ask in the no-blame week
- Which AI tools have you used for work in the last month, and on which device or account?
- What do you use them for? Name two or three tasks.
- Roughly how much time does it save you, or what does it improve?
- Have you ever been unsure whether something was safe to put in?
- What would make it easier to use AI well here?
A line you can send on Monday
“Many of us already use AI tools for work, and that’s understandable: they help, and we haven’t said what’s allowed. That’s on us, not you. This week we’d like to hear what you use and what for, so we can approve a tool that does the same job safely. Nobody will get into trouble for anything you tell us.”
Should you monitor staff or block AI tools?
Not as a first response. Blocking without an approved alternative tends to move use onto personal phones, out of sight, and the ICO expects any monitoring of workers to be necessary, proportionate and explained to them beforehand.
The Information Commissioner’s Office (ICO) guidance Employment practices and data protection: monitoring workers (opens in a new tab) (published October 2023, read 29 October 2026) sets the limits. It says you must:
- be clear about your purpose and choose the least intrusive means to achieve it
- tell workers about monitoring in a way that is accessible and easy to understand
- carry out a data protection impact assessment (DPIA) before monitoring that is likely to be high risk.
It also says covert monitoring is unlikely to be justified in most usual circumstances. The ICO notes on the page that this guidance is under review following the Data (Use and Access) Act 2025, so check the current version before you rely on it. Blocking a website on work devices can be a reasonable control once an approved tool exists. Before then, it mostly teaches people to use their phones.
The one thing not to do
Don’t start with disciplinary warnings or monitoring software. If people are punished for admitting what they use before you’ve offered a safe route, they stop telling you, and you lose the information you need to fix the problem.
What if client or personal data has already gone into a personal AI account?
Tell your data protection lead straight away so they can assess it. Unauthorised disclosure of personal data can be a personal data breach, which must be recorded and, unless it’s unlikely to put people at risk, reported to the ICO within 72 hours.
- Thank the person for telling you. Whether people report problems next time depends on how this conversation goes.
- Write down what went in, when, into which tool and account, and what that account’s settings were (for example, whether chats are used for training).
- Ask your data protection lead or adviser to assess it against the ICO’s personal data breach guidance (opens in a new tab) (read 29 October 2026). Under UK GDPR you must record every personal data breach, and report it to the ICO within 72 hours of becoming aware of it unless it’s unlikely to result in a risk to people’s rights and freedoms.
- Where the tool allows it, delete the chat and check the account’s training setting. Your data protection lead decides whether that is enough.
- Fix the system, not the person: if one person did it, others probably have. Use it to speed up the approved tool and the tramlines.
Separate a good-faith mistake made before you offered an approved route from deliberate misuse after clear rules. The first is a gap in your system. The second is a matter for your normal disciplinary procedure, with HR advice.
How does shadow AI fit with your AI policy and culture?
Shadow AI is usually the first sign that an organisation needs an AI policy. The no-blame week gives you the evidence to write one that fits, and a culture where people say what AI did makes hidden use unnecessary.
Many descriptions of AI adoption start with a stage where people experiment on their own before the organisation sets a direction. In the AI Stairway™ to Success, Scorchsoft’s own model rather than an industry standard, this is step 0, “No formal AI adoption”: people use personal accounts on work tasks, there is no policy, and nobody knows what data has gone where. The next step is approved access, and the one after is people using it well in the open.
- Turn what you learn into a policy with the AI policy template, which starts by asking what people already use.
- Decide who owns AI decisions, and how often you review them, with AI governance for small organisations.
- Once an approved tool is in place, help good habits spread from the people already using it well: see AI champions.
Questions people ask
Is shadow AI illegal?
Using an unapproved AI tool isn’t in itself unlawful. The legal risk comes from what goes into it: personal data handled without a lawful basis or proper safeguards can breach UK GDPR, and confidential client information may breach a contract or professional duty. That’s why the fix focuses on data, not on the tool.
Should we ban ChatGPT and other public AI tools?
A ban with no alternative usually moves use onto personal phones, where you can see even less. Most organisations do better approving one tool on an organisation plan and saying clearly what data may go into it. Product names change, so judge any tool by the trust check rather than its brand.
Can we discipline someone for using AI without permission?
If no approved route or clear rule existed, disciplining someone for hidden use is likely to feel unfair and to drive use further underground. Many organisations offer an amnesty and then apply the normal disciplinary procedure to deliberate breaches of rules that have been clearly communicated. Take HR advice on any individual case.
How do we find out which AI tools staff are using?
Ask them. An anonymous, no-blame questionnaire run over a week, with a written promise that nobody will be disciplined for what they report, tells you more than monitoring software and costs nothing. Share what you learned and what you’re changing because of it.
Does buying a business AI licence solve shadow AI?
It solves half of it. An organisation plan puts the tool under terms you chose, but people also need to know what data may go into it, see colleagues using it openly and have time to learn it. Access without that is often barely used.
Training for the whole team
The AI Stairway™ gives everyone the same careful habits, then helps your team agree its rules and test one AI idea properly. Everyone can start with the free introductory module.
AI training for teamsSources
Every figure on this page comes from one of these. Links open in a new tab.
- Rise in shadow AI tools raising security concerns for UK organisations (survey of 2,003 UK employees by Censuswide) (opens in a new tab). Microsoft UK (vendor research), 13 October 2025.
- British workers spend one billion pounds of their own money on GenAI for work (GenAI Workforce Survey) (opens in a new tab). Deloitte UK (consultancy research; fieldwork by Ipsos UK), 16 September 2026.
- CIPD Good Work Index 2026 (section 5, generative AI) (opens in a new tab). CIPD, July 2026.
- Employment practices and data protection: monitoring workers (opens in a new tab). Information Commissioner’s Office, October 2023 (under review after the Data (Use and Access) Act 2025; read 29 October 2026).
- Personal data breaches: a guide (opens in a new tab). Information Commissioner’s Office, updated 20 August 2025; read 29 October 2026.