Skip to content
Disruptive Lab

Guide AI under the company's control

Shadow AI in Malta and how to control it

Your people already use AI for their work, and the way to bring it under company control is a company AI they prefer, not a ban.

Published
Updated
Reading time
7 min read
Editor
David Lahuta

In short

Most people who use AI at work already use tools they picked themselves. In a 2025 study across 47 countries, 70% of employees who use AI for work used free public tools, and only 42% used tools provided by their employer.1 About half of them said they had uploaded company information, such as financial, sales or customer data, into a public AI tool.1 A ban does not stop this. The same study found such use reported most often where generative AI was banned.1 What works is one company AI that people would rather use, which works with each user's permissions and records every step.

What is shadow AI, and how common is it?

Shadow AI is the AI that people use for their work without the company having chosen it, approved it or even knowing about it. It is usually a free chat tool in a browser or on a personal phone, used with a private account.

It is the normal case, not the exception. In the 2024 Work Trend Index of Microsoft and LinkedIn, a survey of 31,000 knowledge workers in 31 markets, 11 of them in Europe, 75% said they used AI at work and 78% of those users brought their own AI tools.2 The same survey found that 52% of people who use AI at work were reluctant to admit using it for their most important tasks.2

The newer figures point the same way. The University of Melbourne and KPMG surveyed more than 48,000 people in 47 countries, 25 of them in Europe, between November 2024 and January 2025.1 Of the employees, 58% said they use AI at work on purpose and regularly.1 Of those who use it, 70% use free public tools, 42% use tools their employer provides, and 18% pay for public tools themselves.1

Only two in five employees said their organisation had a policy or guidance on generative AI, and almost one in five did not know whether it had one.1

Why do people pick their own tools?

Because they are there, they are free and they help. The study's authors explain the high use of general chat tools by how easy they are to reach, with a plain language interface and no training needed.1

There is also pressure. Almost half of the employees (48%) said they worry about being left behind if they do not use AI at work.1

And in many companies there is simply nothing else. When fewer than half of the people who use AI have a tool from their employer,1 the rest do what any capable person does. They find their own way to get the work done.

None of this is bad intent. The people who use AI on their own are often the most willing to work in a new way. The question for the company is where that work happens and what it takes with it.

What can leave the company?

Whatever people paste into the tool. In practice that is the work itself.

What a person pastes in Why What the company cannot see afterwards
A customer's e-mail thread to draft a reply which names, contacts and details went out
A price calculation or an offer to check it or rewrite it which margins and prices left the company
A contract or a supplier agreement to summarise it which terms and parties were shared
A list of customers or orders to sort it or find patterns whose personal data went where
An internal report to turn it into a presentation what was in the version that was sent

This is not a guess. In the 2025 study, 48% of employees said they had uploaded company information, such as financial, sales or customer information, into a public AI tool.1 Over half (57%) said they had hidden their use of AI or presented work done by AI as their own.1

The security community ranks this risk near the top. The OWASP Top 10 for LLM Applications 2026, the security community's list of the main risks of systems built on AI language models, puts sensitive information disclosure second of ten, and its authors say their confidence is highest there.3 The data it names includes personal and financial data, access credentials and trade secrets.3

The US standards body says the same in its own words. NIST AI 600-1, its risk profile for generative AI, lists data privacy among twelve risks, meaning the leakage and unauthorised use or disclosure of personal or sensitive data.4

For personal data there is also the law. Under GDPR a company may have personal data processed on its behalf only by a provider that gives sufficient guarantees, and only under a contract that says what may be done with that data.5 With a private account that an employee opened on their own, the company has checked no guarantees and signed no contract.

Why do bans fail?

A ban alone does not end the use. The 2025 study found uploads of company information and other use against the rules reported most often by employees whose organisation had banned generative AI (67%), against 33% where there was no policy at all.1 The authors conclude that outright bans may be ineffective and that a policy alone does not guarantee compliance.1

The same study found that 56% of employees had used AI at work without knowing whether it was allowed.1 A rule that people do not know about does not protect anything.

In our reading, a ban mostly moves AI out of sight, for example to a personal phone. The work still goes into a tool, and the company loses its view of it.

What makes AI the company's own?

Four things, and one result that follows from them.

  • One deployment for the whole company. One AI chosen by the company and connected to its systems, documents and e-mail, instead of a tool each person picks on their own.
  • Each user's permissions. The AI works with the rights of the person asking. What that person may not see, the AI does not see either.
  • Data where the company wants it. Data is stored and the AI models run in Europe, or in the company's own infrastructure, and the model provider keeps nothing.
  • Every step recorded. Every question, answer and action is recorded and can be traced later, and changes to data and sensitive steps are approved by a person.

The result is the one that matters. When people have an approved tool that is good for their work, they stop putting company data into private ones.

The security guidance points the same way. NIST suggests that an organisation keeps an inventory of the generative AI systems it uses, keeps a list of approved providers and sets clear rules for acceptable use.4 OWASP recommends checking a user's permission before the AI reads a document, not afterwards, sending an outside model only what the task needs, and enforcing that the provider neither trains on the data nor keeps it in the system itself, not only in a policy text.3

Private accounts One company AI
Who chooses the tool each person the company
What the AI may see whatever the person pastes in what the person's permissions allow
Where the data goes wherever the provider decides where the company decides
What is recorded nothing the company can see every question, answer and action
Who answers for it nobody in particular the company, with a named person

How long does the switch take?

It is not a year-long programme, because nothing that already works has to be replaced. The order matters more than the calendar.

  1. Find out what is used today. Ask people which tools they use and for what, openly and without blame. Otherwise you only hear what they think you want to hear.
  2. Decide what may go where. Which data may an AI see at all, and which only with a person's approval.
  3. Give people the company tool first. Start with what they already do privately, such as writing, summaries and questions over documents.
  4. Connect systems one at a time. Each connection with its permissions and its records, starting where it saves the most time.
  5. Then close the private route, and say why. Once people have something better, the rule is easy to keep.

The first steps take days. Connecting the company's systems follows in small steps, each one in use before the next begins.

Five questions on how AI is used in your company today

  • Which AI tools do our people use for work today, and on whose account?
  • What do they put into them, for example customer data, prices, contracts or drawings?
  • Do we have a written rule on AI, and does everybody know it?
  • Could we show afterwards what was asked and what left the company?
  • Do our people have an approved tool that is at least as good as the one they use privately?

What this means for a company in Europe

For a company in Europe the question is also one of law. GDPR asks the company that is responsible for personal data to have measures in place that ensure, and let it demonstrate, that the data is processed as the regulation requires.5 It also asks for a contract with every provider that processes personal data on the company's behalf.5 Data that an employee pastes into a private account sits outside both. A company AI with data kept in Europe and every step recorded gives the company something it can show.

Next step

If it would be useful, we can meet and look together at how AI is used in your company today, which data goes where, and what the AI Business Harness, one AI for the company under its own control, would need to cover. What you do with it afterwards is entirely your decision.

AI Business Harness

Sources

  1. 1. The University of Melbourne and KPMG, Trust, attitudes and use of artificial intelligence, a global study 2025. Read on 17 November 2026
  2. 2. Microsoft and LinkedIn, 2024 Work Trend Index, AI at work is here. Now comes the hard part. Read on 17 November 2026
  3. 3. OWASP GenAI Security Project, Top 10 for LLM Applications 2026. Read on 17 November 2026
  4. 4. NIST, AI 600-1, Artificial Intelligence Risk Management Framework, Generative Artificial Intelligence Profile. Read on 17 November 2026
  5. 5. Regulation (EU) 2016/679, General Data Protection Regulation, Articles 24 and 28, EUR-Lex. Read on 17 November 2026

Change log

  1. Published.

Let's talk about your project

Leave us your contact details and we will get back to you with a date. Free and without obligation.

With the country code, for example +356.

We use your details only to answer you. Privacy notice

This form is protected from automated submissions by Google reCAPTCHA. Google Privacy Policy (opens in a new window) Google Terms of Service (opens in a new window)

Or contact us directly

Write to us office@disruptivelab.dev
Call us +420 777 702 032

Our preferred language of communication is English.

All contact details and addresses