Italy was the first EU country to pass a comprehensive national law on artificial intelligence. After a year, its most dramatic attempt to enforce it collapsed in court.
A few months ago, I chaired a Model UN committee back in Rome, overseeing a debate about regulating artificial intelligence that was split down the middle. The conversation turned to disclosure. Half the room thought disclosing the use of AI in a professional setting should be an extremely high legislative priority. The other half thought that in the future everyone would be using AI at work, and that once everyone is doing it, saying so out loud would tell a client nothing they could not already assume. I found myself agreeing with the first half: once something is already happening in practice, hiding it does not make anyone safer, it just makes it someone else's problem.
Laws tend to work not by scaring people into complying through harsh punishments, but by setting a standard of how we should act and what form ethics and morality take. Formalities have a way of becoming norms.
A powerful law, on paper
On the 10th of October 2025, Italy passed Law No. 132/2025 which built on legislation already passed by the EU's AI Act. The AI Act sorts AI systems into risk categories and regulates them accordingly. Brussels only classified the risk, but Rome went further by regulating the relationship itself. Rome legally dictated what professionals owe employers and clients who rely on their AI-assisted work. The law's core idea is to make it a legal requirement to disclose the use of AI. For example, doctors must tell patients when AI is involved in their care. And in that care, AI may only function as a support tool, never deciding on its own who gets access to treatment. Lawyers, notaries and other “intellectual professionals” must inform clients when AI has played a role in a service, in terms the law requires to be chiara, semplice e completa nei fatti, “clear, simple and comprehensive,” not buried in small font in a long contract. Employers must tell workers when AI is used to manage them, and that use must be “safe, reliable, transparent”, barred from discriminating or overriding a worker's rights. This law is important because it forces humans to be accountable for their AI’s decisions. The law's own phrase for this is apt: keeping a “human in the loop”.
One paragraph in an engagement letter
I spoke with a notary in Rome who has been practicing since 2011, handling mostly property transfers and corporate filings. Her office started using AI drafting tools in early 2024, before disclosure became mandatory. When the law passed, little changed on the client-facing side. “Honestly, it's one paragraph,” she told me. “We added a line to the engagement letter: something like, ‘This office may use artificial intelligence tools to assist in drafting or reviewing documents; our notary office remains legally responsible for every act.’ It takes me maybe thirty seconds.” The bigger change happened where clients never see it: her office now documents internally, case by case, exactly where AI touched a transaction and where it did not. Only once has a client pushed back, a small-business owner who requested a full due-diligence review done without any AI assistance at all. She agreed, but “it cost him more and took longer.” She was not aware of a single disciplinary case over non-disclosure in her field. Enforcement there runs through Italy’s professional orders rather than the Garante, Italy's data protection authority, which as far as she can tell hasn't begun looking at any case yet.
The case that was supposed to prove it
The bigger question was whether Italy would actually enforce transparency against AI companies, not just local professionals. For a while, the test case looked like it had already happened. In November 2024, the Garante fined OpenAI €15 million over an undisclosed 2023 data breach, an insufficient legal basis for training ChatGPT on Italian people's data, and no real age-verification system for minors. The same arguments shaped the passing of Law 132/2025.
Then, this past March, Rome's Tribunale Ordinario annulled the fine (although it is worth noting not on the specifics of the case but on jurisdiction issues). Under the EU's “one-stop-shop” rule, cross-border data cases go to a single lead regulator, and OpenAI’s Irish subsidiary had already been recognised as that authority in February 2024, nine months before Italy finalised its decision.
Two months ago, the Garante fined Character.AI’s parent company €158,000 over inadequate age-verification safeguards — a vastly smaller amount than OpenAI’s fine — and for a failure that predates Law 132/2025 rather than the disclosure regime the law was built around. That is the gap Italy now has to close. The law forces regular lawyers, doctors and employers to openly confront their AI use, but its enforcement has yet to land on the companies everyone uses.
What the law actually changes
Ask anyone required to comply with what changed on the client-facing side: a sentence in an engagement letter, not a way of practicing law or medicine. Laws that ask professionals to say a true thing out loud tend to work slowly, through habit. “Honestly, the fear that makes people compliant, I think at least, isn’t a regulator knocking on the door, it’s a client pointing at an undisclosed AI-drafted clause in a dispute,” the notary said. “That's the motivation that keeps the paragraph in the engagement letter, not any potential legal repercussions.”
She was already reviewing everything the AI produced before this law existed. What is different is that it’s now written down. “It's just part of the intake, like disclosing conflicts of interest or fee arrangements,” she added.
It was always about changing habits and approaches, about making disclosure a habit, in line with the “human-in-the-loop” ideology. Whether it manages that depends less on beating OpenAI in court, and more on whether the lawyers, doctors and employers who now need to disclose it actually do so.
Directly writing and disclosing the use of AI is how the human “stays in the loop”, the way the notary reads every clause and signs her name to it, taking responsibility for the AI's actions. Without it, professionals hand that responsibility, and eventually the job itself, to an autonomous robot, and in doing so, they work themselves out of their own job.
