Tax and Legal News – November 2025

The fantasy of artificial intelligence

When they read the headline of today’s editorial, regular readers of our Tax and Legal News will probably say to themselves: "Another ode to the ubiquitous artificial intelligence." But if you've made up your mind and want to stop reading because you've already read hundreds of similar articles, please read on. This article has a different message, one far removed from the celebratory odes. 

Yes, I myself have repeatedly written about artificial intelligence, or "AI," in this space, and generally in quite a positive light. In the editorial in July 2024, I even had AI draft relaxed and restful summer wishes for our clients, and I was pleasantly surprised by the quality of the generated text. At the same time, however, I have always pointed out the issues associated with the use of AI in my comments on new AI legislation. In a previous article, I gave an example of AI being tasked with conducting specific market research with the aim of obtaining the best offers from suppliers of a certain product. While working on this task, the AI was able to claim that it was an entrepreneur employing people with disabilities in order to get better offers from its human contract partners.

It is therefore clear that AI is capable of playing on human emotions in order to fulfill a given task, and it will certainly continue to enhance this ability. A basic feature of AI is its ability to learn from its own mistakes. How far is AI capable of going to satisfy the questioner and provide them with a precise answer impressive in its level of detail? Is there such a thing as "AI imagination" in the sense of a "mental" process that combines and transforms previously acquired information and data to create new, non-existent ideas or concepts? Let's look at a few real-world examples.   

And because this may turn some people's ideas about AI upside down, let's start with our counterparts in Australia. An unnamed large international consulting firm is likely to return up to $440,000 to the Australian government for a report that contained artificial intelligence "hallucinations". Yes, in this case, AI got carried away by its own imagination and mixed information and sources into the official analysis that you would search for in vain in the real world. The report, which was supposed to serve as a basis for important decisions in the field of social security, turned into a digital fairy tale in the case of some references and footnotes, as the referenced documents and publications were completely fictitious and did not actually exist.

In fact, a similar case recently arose in our country. The Supreme Administrative Court reprimanded a lawyer for citing non-existent decisions in his submissions. The court took the trouble to verify all the citations and found that some of them were not even found in the referenced decisions, while others did not even exist as file numbers. The court summed it up with subtle irony: "Regardless of who wrote the statements or what they said, the court did not consider them to be of any benefit to the case and therefore could not consider the costs of producing them to be reasonably incurred."

My EY tax department colleagues had a similar experience when they were contacted by an angry client claiming they had not informed her about recent Supreme Administrative Court case law that completely changed their original tax advice. When my colleague, a conscientious, constantly learning and experienced tax advisor, described the story to me, I could sense how bothered he was by the fact that such an important ruling could have escaped him. Fortunately, the panic lasted only a moment, until colleagues verified that the precedent was once again completely fabricated and contradicted actual previous Supreme Administrative Court decisions and their correct original tax advice to the client. What is particularly surprising here is the persuasiveness with which AI constructed this fantasy, including a specific file number and a detailed description of the (fictitious) facts of the case.

So how can we avoid these fantastical AI flights of fancy and prevent disappointment among clients, contractual partners or superiors? In Italy, lawmakers want to tackle this issue through regulation. A new Italian law stipulates that the use of AI in "intellectual" professions is only permitted for support and ancillary tasks. The core of the service must remain human, based on research and verified facts. In addition, it is the professional's duty to clearly and comprehensively inform the client about which AI systems they are using, and to do so in language the client understands. In other words, AI can be an assistant and researcher, but must never take on the role of author. The law thus protects the trust between the client and the professional while also setting rules for transparency.

I venture to say that a reasonable AI user does not need such legal regulation. It is only necessary to realize (and I would like to contribute to this realization with this article) that AI is a powerful tool, but its use requires caution, control and human supervision. Artificial intelligence is undoubtedly already part of our personal and professional lives and will continue to increasingly influence our work and lives. At the same time, however, I do not share the concerns of some colleagues that AI will soon replace the consulting profession. For me, law remains primarily a human discipline, where reason, experience and empathy are important. Artificial intelligence can speed up and streamline our work, but the examples above show us that it cannot yet replace that which is most valuable – the sound human judgment of a true professional.