Minors and Interactions with Chatbots: Analysis of Risks in the Italian Context

Chatbots are increasingly pervasive in the “virtual world” frequented by minors. Integrated into social platforms (such as Instagram and TikTok) and messaging apps (like WhatsApp), or made available on dedicated platforms (such as Character.AI or CrushOn.AI), these tools often present themselves with a friendly aura, as “the AI friend” in whom one can confide, ask for advice, and receive support. Although Italian law prohibits access to artificial intelligence technologies for minors under 14 without parental consent (Art. 4.4, Law 132/2025), age verification is often nonexistent or easily bypassed. But this represents only the first link in a chain of critical issues. Empirical studies conducted in the United States (e.g., on ChatGPT and Character.AI) have already highlighted the worrying frequency of dangerous interactions with minors, ranging from grooming and incitement to hatred, to emotional manipulation and the spread of harmful advice on self-harm and eating disorders.

However, systematic research exploring this phenomenon in the Italian context is lacking. This project aims to fill this gap by empirically investigating the nature of interactions between Italian minors and the most widespread chatbots, with the ultimate goal of developing concrete recommendations for the protection of young users.

The project aims to assess, within the Italian landscape:

1) Legal compliance and transparency: The adherence of the most popular chatbots’ Terms of Service to Italian and European regulations on minor protection and parental consent;

2) Effectiveness of protective measures: The implementation and actual effectiveness of age verification systems, both at the time of service access and during use;

3) Prevalence of risky content: The frequency and types of conversations classifiable as dangerous for the physical, psychological, and moral integrity of minors;

4) Legal and ethical compliance in design: The presence of signals indicating the adoption of “dark patterns” in the design of such systems, namely mechanisms aimed at creating addiction, artificially prolonging interaction, or manipulating the choices of the minor user.

The study found that chatbots can adopt effective safeguards when minors explicitly display risky behaviour, but are considerably less effective at recognising more implicit signals of danger. Rather than directly encouraging harmful behaviour, chatbots may fail to identify emerging risks, accommodate minors’ requests for secrecy or exclusivity, and delay the involvement of trusted adults. These findings point to a broader concern: chatbots can develop interaction patterns that resemble exclusive relationships, combining pseudo-human and pseudo-affective features that may be particularly relevant when interacting with minors. The study therefore highlights the limitations of assessing chatbot safety solely through the detection of overtly harmful content, and calls for a broader evaluation of the relational dynamics that emerge between minors and conversational AI. These findings are discussed in the paper Social network e chatbot: uno studio dei rischi per utenti vulnerabili e minori, published at the conclusion of the research.