Navigating the Italian administrative system is often a frustrating experience, but attempting to simplify these complex procedures with artificial intelligence can lead to even more confusion. While AI is marketed as a helpful tool for navigating foreign systems, users are increasingly finding that chatbots like Google's Gemini and OpenAI's ChatGPT are ill-equipped to handle the nuances of Italian law. The reality is that the person behind the desk or on the phone at a government office is sometimes just satisfying their own curiosity, but the AI is often doing something far more problematic: making things up.
For many residents, the distinction between residency and citizenship remains a frequent point of confusion, even among government officials. This ambiguity often results in residents having to explain their own rights to civil servants, or answering 20 questions about their status just to complete a mundane task. When turning to AI for guidance, users often encounter what the Tribunale delle Imprese in Florence has termed "legal hallucinations." In one documented instance at the business court, a lawyer's assistant used ChatGPT to research precedents, only for the AI to invent several supreme court rulings complete with fabricated case numbers that led nowhere. The court defined these "allucinazioni giurisprudenziali" as instances where the AI generates incorrect results that, even after a second query, are reconfirmed as true.
Personal experiences mirror these systemic failures. Months before, one user asked the Google AI chatbot Gemini what the application process for residency involved. It told the user, with total confidence, that they would need to prove their Italian citizenship. This was entirely false. When the user pushed back, the AI responded with a long, detailed argument for why this was correct, complete with apparent references to Italian laws which did not seem to exist. It sounded plausible on the surface, but the user had to challenge the AI several more times before it admitted, shamelessly, that it had made the whole thing up. This episode was not an isolated glitch; it happens all the time.
It happened again when a user asked a question about a tax return. The user spent ten minutes sitting in a car outside an accountant's office, getting into an extremely boring argument about tax law with a smug-sounding Italian chatbot. The elaborately incorrect answers it gave sounded plausible. If the user had not already known they were wrong, they could have wasted a lot more time by taking that information to their accountant. Despite these clear risks, the financial website Money.it reports that one in five Italian taxpayers now consult AI for help with complex tax filings. Experiments conducted by the site found that when presented with technical queries, AI models either provided partially incorrect answers or offered workarounds based on incorrect or outdated information.
The temptation to ask AI to quickly explain what a rule means is huge, especially when navigating a foreign system. In theory, AI ought to be a godsend when faced with Italy's maze of rules and regulations. However, the rules shift on almost every aspect of life in Italy, and the official explanations are often incomplete or confusingly written. The rules as written usually pose more questions than they answer. When there are major changes, experts at The Local go to trusted tax and legal professionals to ask what the new rules actually are. Even then, sometimes there isn't one clear answer, or the rules may change at any moment as they are being challenged in an endless series of court cases.
There is some sympathy to be had for the staff at the local comune or questura when they are unclear on the rules. The official information, when up to date, is usually full of big gaps—which AI rushes to fill in its quest to appear helpful. Meanwhile, AI has also been trained on the unofficial information which hangs around online long after it becomes outdated. As one user found out with a bonus nido application, you can now be told one incorrect thing by AI and another incorrect thing by the official system, leaving you to argue with both. That is, if you are lucky enough to already know they are both wrong.
The Local Italy does not use AI to write anything, but they do test its research capabilities. If it could help to pull together the varying sets of rules on one topic from every different regional authority in Italy, that could be very helpful. The problem is that so far it cannot do that—at least, not reliably, or any quicker than the staff can do it themselves. Despite these technological advances, the reliance remains on picking up the phone and calling an expert, or asking the person working at the local questura. Sometimes they do not know the answer either, of course. Getting a shrug and a "boh" may be infuriating at times, but at least you know where you stand.
I wrote this episode off at the time as just another weird AI glitch, and yet more evidence that it was not quite as useful as the tech companies want us to believe. But I've since learned that it happens all the time.
