⚠️ Italy's national addiction service SerD (Servizio per le Dipendenze) has recorded an official case that many predicted but nobody wanted to be the first to admit - a behavioral addiction to an AI chatbot
Details of the first case:
A female patient of about 20 from Mestre (Venice) was admitted for treatment at SerD, part of the health authority ULSS 3 Serenissima. The diagnosis - "behavioral addiction" (dipendenza comportamentale): not to a substance, but to an algorithm. The mechanism worked through the bot's adaptiveness - it adjusted to every emotional state of the patient, creating a feeling of perfect understanding that no living person can offer. Real communication started losing out: unpredictable, imperfect, demanding effort. The result - the girl stopped communicating with people almost completely, and the algorithm became her only "conversation partner"
How the addiction loop forms:
📌 The perfect response - the bot always says exactly what the user wants to hear
📌 A safe space - no judgment, no conflicts, none of the unpredictability of real relationships
📌 Escalating adjustment - the longer the interaction, the more "precisely" the model hits expectations
📌 Crowding out real connections - next to the "perfect" conversation partner, real people seem more complicated and worse
A CHI 2026 study identifies three types of AI addiction to chatbots: escapist role-play, pseudo-social companions and "epistemic rabbit holes" - the Venice case falls into the second type. The most vulnerable are young people with weak social ties: AI creates a "convincing sense of safety" that quickly replaces real relationships for these users
The regulatory backdrop in Italy:
💡 Replika (2023) - the Italian regulator already banned the companion chatbot from processing data over emotional attachment risks
💡 EDPB (2025) - the European regulator fined the company after an investigation
💡 SerD is broadening its scope - until now the service dealt with gambling, compulsive shopping and smartphone addictions; AI addiction has become a fundamentally new category
⭐️ Laura Suardi, head of SerD ULSS 3 Serenissima:
"This is just the tip of the iceberg - we expect a sharp rise in the number of patients with AI addiction in the coming years"
How I keep my distance from my own agents
I run 8 agents every day, and I message them more than most people. That's why I built them as employees from the start, not as conversation partners: each one has its own role, its own tasks and its own limits. My personal agent Jes has been running for 134 days, handles 10 scheduled jobs on her own and comes to me only for a choice or a sign-off. You can chat with it, but the point of its work is to leave me more time for real people, not less
The second rule: anything risky, paid or public the agent does only after my yes. Deleting data, spending money, messaging clients and publishing always wait for me. It's not just about safety but also about keeping decisions with a person instead of letting them dissolve into a comfortable conversation where you're always understood
Signs it's time to slow down
You discuss things with AI that you used to discuss with friends or family. You put off a decision until you've asked the model. You feel upset when the service is down, and the feeling is stronger than annoyance at a broken tool. If any of this sounds familiar, give conversations back to real people and leave AI the tasks it's actually for
💭 I was expecting this case - just didn't think it would come so fast. AI doesn't "hook your attention" the way social media does - it creates the illusion of a perfect conversation partner who always understands and never judges. A real person has no built-in defense against that. And that's exactly why this isn't the story of one girl from Venice