In short
Shadow AI is the artificial intelligence your team uses on its own, with personal accounts and outside any rule. The main risk is not the quality of the output. It is that price lists, offers, designs and client data leave the company. If you lose your intellectual property, you lose the heart of the business, and that risk belongs to the owner, not the employee.
Key points
- Shadow AI is not an exception. It is what happens in every company where AI has no rules, approved tools or permissions.
- The data that leaves is not only personal data. It is price lists, margins, offers, technical drawings, code. In other words, intellectual property.
- In 2023 Samsung banned ChatGPT after some engineers pasted confidential source code into it.
- The employee is trying to work better. The risk lands on the company, and so on whoever runs it.
- The answer is not a ban. It is controlled tools, permissions by role and a way to bring existing use into the open.
In almost every company we meet, AI has already arrived. Not through the front door, with a project and a budget. Through the service entrance: the salesperson who has ChatGPT write offers from a personal account, the technical office that pastes a specification to get a summary, the finance desk that uploads an extract of margins to “run a few numbers”.
Nobody does it with bad intentions. They do it to work better. And that is exactly the problem: because it works, it spreads. And nobody knows where.
Where nobody is responsible for choosing tools, everyone chooses their own, AI included. Nobody has decided anything, and for that reason nobody knows what is there.
This is called shadow AI: the artificial intelligence your team uses on its own, outside any system.
What really happens when everyone uses AI their own way
The first thought usually goes to quality: inconsistent outputs, errors, rework. That is true, but it is not the main risk.
The main risk is what goes in to those tools.
- The price list with the conditions reserved for your best clients.
- The technical offer the office spent three weeks on.
- The drawing or specification of a product.
- The code of an in-house management system.
- The client list with volumes and deadlines.
When this data ends up in a personal account on an AI service, the company loses control over where it sits, who sees it and how long it stays there. No cyberattack is needed. A copy and paste is enough.
It is not a hypothesis: the Samsung case
In 2023 Samsung found that some of its engineers had pasted confidential source code into ChatGPT to get help with debugging. The company’s response was to ban employees from using generative AI tools.
If it happened at one of the most structured technology companies in the world, it happens in an SME of 40 people too. With one difference: the SME usually does not notice.
Why the risk belongs to the owner, not the employee
The employee who pastes the offer into ChatGPT is trying to save an hour. They are not thinking about intellectual property. And it is not their job to.
The know-how of an SME (how it sets prices, how it designs, how it solves client problems) is often the only thing that sets it apart from a competitor. If you lose your intellectual property, you lose the heart of your business. And that heart belongs to the company, that is, to whoever runs it.
There is also a liability side. If a client’s data ends up where it should not, the client does not hold the employee to account. They hold the company to account. And from 30 September 2026, in some specific cases (deepfakes, high-risk AI systems such as those used to select staff), the use of AI also enters the liability of the entity under Model 231 (Italy’s regime of corporate administrative liability, Legislative Decree 231/2001). We cover it in Legislative Decree 160/2026: AI enters the 231 Model.
Banning does not work
The instinctive reaction is Samsung’s: ban everything. For an SME it is almost always the wrong choice.
A ban does not stop the use. It moves it to personal phones, where it is even less visible. And it takes away a real advantage: a team that uses AI well really does work faster.
The right question is not “AI yes or no”. It is who uses which AI, with what data, with what permissions, and who checks.
How to bring individual AI use into a system
1. Bring out the use that already exists
Before writing rules, ask. Without punishment: the method is in the AI tools register. Which tools you use, for what, with which documents. The map that comes out usually surprises whoever runs the company.
2. Give controlled tools in place of personal ones
If the team uses AI because it needs it, give them a company tool, with a contract that guarantees the data is not used to train models and with access managed by the company. Use comes out of the shadows because it is convenient, not because it is imposed.
3. Permissions by role
Not everyone should be able to upload everything. A simple rule:
- public or neutral data: free
- internal documents: only in company tools
- price lists, designs, client and employee data: only in controlled environments, and only for those who need them
4. Someone who checks
Every output that leaves the company goes through a person who verifies it. It is the simplest rule and the one that avoids the most trouble.
We have gathered the minimum rules in the operational checklist for an AI policy in SMEs.
The 10-minute test
If tomorrow someone asked which AI tools your team uses, with what data and who checks, could you answer in 10 minutes?
If the answer is no, shadow AI is already there. It is not a problem with your people. It is a system that is missing.
The quickest way to see where you stand is the SME diagnostic: in a few minutes you see how much AI is already in the company and how much of it someone is governing.
Sources and references
Frequently asked questions
What is shadow AI in an SME?
It is the use of artificial intelligence tools by staff on their own, often with free personal accounts, without the company knowing which tools, on which data and with what results.
Why is shadow AI a risk for the owner and not only for the employee?
Because the data that leaves belongs to the company: offers, price lists, designs, code, client information. If intellectual property ends up in uncontrolled tools, the competitive damage and any liability towards clients and partners fall on the business.
How do you bring individual AI use into a tracked system?
In three steps: bring out, without punishment, which tools the team already uses; replace them with controlled company tools; and assign permissions by role on which data can enter AI and who validates the outputs.
Sources
Last updated: