Shadow AI at Work: The Risks and How to Handle Them

Shadow AI is the use of AI tools at work that the company has not approved, does not know about or cannot see. An employee pastes a customer complaint into a free chatbot to draft a reply, a developer asks a personal AI account to fix production code, a manager uploads a salary spreadsheet to “make a chart”. None of them mean harm. Most are trying to do their job faster. But each one may send company data somewhere nobody has checked.

This guide explains why shadow AI happens, what it actually puts at risk, how to find out how much of it you have without turning it into a disciplinary hunt, and how to replace it with something people will genuinely prefer to use.

What counts as shadow AI

The term borrows from shadow IT, the older problem of staff using unapproved software and cloud services. Shadow AI is narrower and, in some ways, riskier, because the whole point of an AI tool is that you give it your content.

Typical examples include:

• personal accounts on consumer AI chat apps used for work tasks;
• browser extensions that read every page and form an employee opens;
• AI features switched on inside other software without anyone reviewing them;
• meeting recorders and transcription bots joining calls on an individual’s say-so;
• free online “AI converters” for PDFs, images or spreadsheets;
• scripts that send company data to an AI API using a personal key.

What makes something shadow AI is not the tool itself. A tool can be perfectly good. It is the lack of a decision: nobody checked where the data goes, who can access it, whether it is used for training and what happens when the employee leaves.

Why shadow AI happens

Banning a useful thing rarely makes it disappear. Unapproved AI use usually grows for a few predictable reasons.

• There is no approved option. If the company offers nothing, people use what they have at home.
• The approved option is worse. A locked-down tool with a slow sign-up process loses to a free app that works in ten seconds.
• Nobody said anything. Without a policy, most staff assume that if a tool is free and popular, it must be fine.
• Pressure to deliver. When workloads are high, people reach for whatever saves an hour, especially for writing and summarising.

Understanding this matters, because it points to the fix. Shadow AI is usually a demand signal, not a discipline problem.

The real shadow AI risks

It is easy to exaggerate shadow AI risks into a vague fear. It is more useful to be specific about what can actually go wrong.

Confidential data leaving your control

Customer records, contracts, source code, financial figures and HR files pasted into a tool you have no agreement with. Depending on the service and the account type, that content may be stored, reviewed or used to improve models. You often cannot get it deleted, because you are not the customer, the employee is.

Personal data and GDPR

If the content includes personal data of customers or staff, the company is still the controller under GDPR. Sending it to a processor you have no contract with, possibly outside the EU, is a compliance problem in its own right. Article 32 of the GDPR requires appropriate security measures, and an unknown chatbot is hard to defend as one.

Lost work and no audit trail

When an employee leaves, so do their chats, prompts and drafts. If a decision was based on an AI summary, there is no record of what the AI was given or what it said.

Quality and accountability

AI output that goes to customers without anyone knowing it was AI-assisted cannot be checked properly. Errors, invented facts and inconsistent tone reach clients with the company’s name on them.

Security exposure

Browser extensions and third-party integrations can request wide permissions. Unreviewed connectors to email or file storage are a particular concern, as are AI tools that can be manipulated by malicious content, a risk explained in prompt injection explained.

Shadow AI vs approved AI compared

Question	Shadow AI (personal, unreviewed)	Approved company AI
Who has the contract?	The employee, or nobody	The company
Is data used for training?	Often unknown	Checked before approval
Where is data stored?	Often unknown	Known and documented
Can the company delete or export it?	No	Yes, if the terms allow
What happens when someone leaves?	Work leaves with them	Shared projects stay with the team
Is usage visible?	No	Yes, at least at account level
Cost control	Scattered personal subscriptions	One budget

How to find shadow AI without a witch-hunt

The quickest way to drive unapproved AI use further underground is to treat discovery as an investigation. People will stop telling you what they use, and you will lose the information you need.

Ask openly, with an amnesty

Run a short, anonymous survey: which AI tools do you use for work, for what tasks, and what would you need from a company tool? Say clearly that nobody will be punished for past use. The answers tell you both the risk and the demand.

Look at the obvious signals

Expense claims and card statements for AI subscriptions, browser extension lists on managed devices, and third-party apps connected to company email and file storage. Your IT provider can usually report these without reading anyone’s messages.

Talk to the heavy users

The people already using AI well are your best source for what works. Make them part of choosing the approved tool rather than the subject of a crackdown.

A practical plan for handling shadow AI

For a small or mid-sized company, a sensible response fits on one page.

1. Write a short AI use policy. What tools are approved, what data may never be pasted into any AI, who to ask. Our guide on how to write an AI use policy walks through it.
2. Choose at least one approved assistant. Check the vendor’s terms on training, storage location, deletion and access. The AI privacy checklist lists the questions to ask.
3. Make the approved option easy. Quick sign-up, shared projects for common tasks, enough usage that people do not hit limits on day one.
4. Define the red lines. For example: no passwords, no full customer databases, no health or HR files, no unreleased financials in any AI tool, approved or not.
5. Train briefly and repeat. Under the EU AI Act, organisations that deploy AI systems are expected to take measures to ensure sufficient AI literacy among staff; Article 4 sets out that obligation. A one-hour session with real examples from your own work is a good start.
6. Review every six months. Tools change quickly. Re-run the survey and update the approved list.

For larger organisations, frameworks such as the NIST AI Risk Management Framework provide a fuller structure, but most small businesses do not need to start there.

Replacing shadow AI with an approved assistant

The goal is not to stop people using AI. It is to move that use somewhere the company can see and stand behind. That only works if the approved tool is genuinely good.

Ask Mio is one option built for this situation. It covers the jobs people usually turn to unapproved tools for, chat, writing, research with cited sources, code and design, in one assistant, so teams do not need five separate tools. Chats and files belong to the user and can be exported or deleted at any time, data is not used to train models, and the servers are in the EU, in Germany. The Business plan includes 5 team seats, team and shared projects and API access, so shared work stays with the company rather than in someone’s personal account.

It will not fit every organisation. A large enterprise with strict procurement may need contractual terms, single sign-on or audit features that a small-team product does not offer, and should check those requirements against any vendor. For background on hosting and data protection, see EU-hosted AI and GDPR.

Signs your approved tool is not working

Approving a tool is not the end of the job. If people keep drifting back to personal apps, the approved option is missing something, and it is worth finding out what.

• Low adoption after a month. If few people have logged in, ask why. The answer is often a slow sign-up, a confusing interface or a missing feature.
• People hitting limits. Usage caps that run out mid-week push staff back to free tools. Check whether the plan matches real usage; in Ask Mio, extra point packs work immediately and the Business plan can refill points automatically.
• Tasks the tool cannot do. If the marketing team needs images and the approved assistant has no design mode, they will find one elsewhere.
• Work still living in personal accounts. If drafts and research are not appearing in shared projects, the habit has not changed yet.

Treat these as product feedback. Fix the gap, tell people what changed and check again in a few weeks.

What employees can do today

If you are an employee and suspect you use AI tools your company has not approved, you do not need to wait for a policy.

• Remove names, account numbers and other identifying details before pasting anything into a personal AI tool.
• Never paste passwords, API keys or credentials, even to “fix” a configuration.
• Tell your manager what you use and why. You are probably not the only one, and it helps the company choose a proper tool.
• Keep a copy of important AI-assisted work in company systems, not only in a chat history.

Our guide on what not to share with AI has a fuller list.

Frequently Asked Questions

What is shadow AI?

Shadow AI is the use of AI tools for work without the organisation’s knowledge or approval: personal chatbot accounts, browser extensions, transcription bots or AI features nobody reviewed. The tools themselves may be fine. The problem is that nobody has checked where company data goes, whether it is used for training, who can access it and what happens to it when the employee leaves.

Is shadow AI illegal?

Using an unapproved tool is not illegal in itself, but what goes into it can be. Sending customers’ or employees’ personal data to a service the company has no agreement with can breach GDPR obligations, and confidential client material may be covered by contracts or professional secrecy rules. The legal exposure usually sits with the company, not only with the individual employee.

Should we just ban AI tools at work?

Bans rarely work on their own. If AI saves people real time and there is no approved alternative, usage tends to move to personal phones and home accounts, where it is even less visible. A better approach is a short policy, clear red lines on sensitive data and at least one approved assistant that is easy and pleasant to use.

How do I find out which AI tools my staff use?

Start with an anonymous survey and a promise of no penalties for past use. Then check expense claims for AI subscriptions, browser extensions on managed devices and third-party apps connected to company email and storage. Talk to the people who already use AI heavily; they can tell you what works and what an approved tool needs to offer.

What should an approved AI assistant offer to replace shadow AI?

Clear terms that your data is not used for training, a known storage location, the ability to export and delete data, shared workspaces so work stays with the team, and enough capability that people do not miss their personal tools. Ask Mio, for example, runs on EU servers, does not train on user data and offers team seats and shared projects on its Business plan.

The Bottom Line

Shadow AI is mostly a sign that people want AI help and have not been given a good way to get it. Treat it as demand, not misconduct: find out what people use, set a few clear red lines and give them an approved assistant that is easier than the alternatives. That protects your data and keeps the productivity gains. If you want an EU-hosted assistant for your team that does not train on your data, compare Ask Mio’s plans and start with the free one.