AI is already working in Polish companies. The problem is that often “silently”

AI has become loud again, or rather it has been loud since the moment artificial intelligence left the laboratories and entered everyday life for good. However, recently the topic has been coming back with a vengeance. Reason? More responsibilities are coming for companies using AI.

From August 2, the provisions of the AI ​​Act regarding the transparency of artificial intelligence systems will come into force in the European Union. In practice, this means, among other things, that the user should know when he is talking to a machine, and content created or modified by AI – including deepfakes – will have to be marked appropriately.

Sounds like another EU obligation to check off? Not necessarily. Because artificial intelligence is already working in many companies – often faster than any rules for its use have been created. Employees use AI tools to create offers, analyze information or prepare materials, but organizations do not always know what this practice looks like and what risks it carries.

I talk to Mikołaj Motel about what will really change in companies from August 2, who is responsible for artificial intelligence errors, how to safely use new tools and why Polish business is quickly adopting AI on the one hand, and still does not always keep up with regulations on the other.

Beata Anna Święcicka, “Wprost”: We are talking just before the next version of the AI ​​Act is released. Artificial intelligence already works in many companies, but it is not always clear who uses it and how. Do entrepreneurs realize what tools their employees use and what responsibilities this entails?

Nicholas Motel, 30Elevate: Artificial intelligence is working at full capacity in many Polish companies today, but no one has employed it. The salesperson creates offers using a model, marketing generates graphics, someone from the administration checks the interpretation of regulations in the chat. No one is hiding anything, and no one asked about it. The phenomenon already has a name: shadow AI.

This will change soon. The next version of the AI ​​Act will come into force in Polish companies.

Yes, from August 2, it ceases to be an internal matter of the company. On this day, the main part of the EU regulation on artificial intelligence comes into force. We’ll probably see in the headlines that “penalties are starting.” This is not true, sanctions for prohibited practices have been in force since August 2025.

August 2026 brings something different, and for most companies, more important: the obligation to be transparent. Anyone who uses AI chat in customer service is expected to tell the interlocutor that they are talking to a machine.

So the biggest challenge for companies will not be the use of artificial intelligence itself, but organizing what is already happening inside the organization?

You need to immediately cut off a misunderstanding that goes the other way, because it repeats itself just as often. The mere fact of using AI does not oblige the company to label every content. A product description, offer or email to a customer written using the model does not require a label. The provision targets situations in which the recipient may mistake the machine for a human.

Poland is a country famous for its speed in adapting new technologies. What is the scale of AI use by employees in Poland?

The data emerges with a picture that should please us. In the SD Worx study from July this year, on a sample of 16.5 thousand employees from sixteen European countries, 38.4 percent of Poles declare that they regularly use AI tools at work. The European average is 29.3 percent. We are not in the tail, we are in front of the peloton.

So where is the problem?

The problem is not the interest in AI itself, but how companies approach its use. The June Cyberportrait of Polish business, prepared by ESET and DAGMA on a sample of 1,026 people, shows that formal rules for the use of AI apply to 27 percent of respondents.

Moreover, over one third declared that they would try to bypass the company’s blockade of such tools, for example by generating content on private equipment and sending it to a work mailbox. This is not a rebellion. This is an employee who has a task due yesterday and a tool that cuts it in half.

There is an obligation that is least talked about, although it has been in force for a year and a half. The regulation requires companies to support AI competences among people who use these tools on their behalf. There is no form to fill out and no office to report it to. There is an expectation that people understand the tool they are working with.

What about liability?

This is the most costly misunderstanding: whose responsibility it is. Air Canada showed it. The carrier’s chatbot assured the passenger that he could collect the bereavement discount after the trip, and when the airline refused to pay, it defended itself before the tribunal with an argument worth remembering: that the chatbot is a separate entity, responsible for its own actions.

A Canadian tribunal rejected this in February 2024 and ruled that the company is responsible for everything on its site, regardless of whether a human or program wrote it.

It’s the same with us, and since August it has been an obligation under the regulation. It is the responsibility of the company using the tool, not the model provider. OpenAI, Google and Anthropic have their own separate commitments. I build AI chats for companies and I know how little it takes for a model to say one sentence too many. If such a chat pretends to be human or promises the customer something that is not included in the price list, pointing the finger at the supplier will not help. The work can be outsourced to a model. No responsibility.

So what should a company do before entering new obligations? Where to start organizing the rules for using AI?

It is worth starting a conversation in a company not with a document, but with three questions. What AI tools do my people use? What data cannot be pasted into them. Who checks the result before going to the client. The answers usually take one afternoon and another afternoon to write them down.

There is one more number left, the most important for me. Two-thirds of those surveyed in Poland by SD Worx believe that AI systems should operate transparently. The obligation that enters into force on August 2 is not a whim of Brussels despite the people. It’s late compared to what these people already expect from companies.

Thank you for the interview.

BIO: Mikołaj Motel, co-founder of 30Elevate. Designs websites and implements AI agents in small and medium-sized companies. 30elevate.com.