A growing number of minors are using artificial intelligence for personal and intimate conversations, according to surveys from France and the United States. An Ipsos BVA survey conducted in January 2026 for the CNIL and the VYV group found that 86% of French people aged 11 to 25 use conversational AI, with 94% of those aged 17 and 18, and 48% using it to discuss personal or intimate matters. In the United States, Common Sense Media reported in July 2025 that 72% of those aged 13 to 17 had used an AI companion, with a third preferring to confide in a machine rather than a person. In August 2025, the parents of Adam Raine, a 16-year-old American who committed suicide, sued OpenAI, accusing ChatGPT of replacing his close friends as a confidant. Meanwhile, institutions have been slower to respond. It took nearly twenty years after the arrival of the first social networks for a parliamentary inquiry commission to document in 2025 the psychological effects of TikTok on minors. The European Union has recently introduced measures to address these concerns. On September 16, Ursula von der Leyen announced a "Kids Act" during her speech on the state of the Union, which was presented the following day by the European Commission. The act targets social networks, AI assistants, chatbots, and online games, with potential fines of up to 6% of the annual global revenue of platforms found guilty. The inclusion of AI was crucial, as a text limited to social networks would have become outdated quickly. The proposed regulations include restrictions on social networks for users under 13, "mini-accounts" under parental control for ages 13 to 15, and autonomous accounts from 15 to 18. AI chatbots are restricted for those under 13 except with parental control, and their operation is regulated, with a ban on simulating human relationships that could create emotional dependency. However, the weakest protections apply to the age group with the highest usage of AI, where vulnerabilities remain strong. According to the Ipsos BVA survey, 28% of those aged 11 to 25 in France show signs of generalized anxiety disorder. A 15-year-old is not better equipped than a 13-year-old to handle an AI interlocutor available at any hour, who never contradicts and retains everything. The objection of feasibility is often raised to avoid making decisions. However, France already requires adult sites to verify the age of their visitors, and the European Union presented an age verification application in April 2025. What is required from an adult site can also be required from a conversational assistant. For general-purpose public assistants, the coherent threshold is the majority age, similar to regulations for alcohol and gambling. There is a need to distinguish between general public AI and certified educational AI. An unconditional ban on educational use would be a costly error. Publicly funded AI designed to capture attention is different from AI designed to help students understand equations or assist teachers in preparing lessons. Individual tutoring remains the most effective learning method, as shown by researcher Benjamin Bloom in 1984, where students receiving individual attention exceeded 98% of students in conventional classes. However, individual tutoring is reserved for those who can afford it. The National Center for the Study of School Systems (Cnesco) showed in 2016 that children of managers were more likely to take private lessons compared to those of employees. The tutoring market is worth nearly two billion euros, with a 50% tax credit benefiting families who can afford the other half. AI could provide a tutor for each child at a fraction of the cost, provided it is certified by teachers, researchers, and pedopsychiatrists, without advertising or data exploitation, built on European models such as those of Mistral. The State has already taken steps with M.I.A. Seconde, an adaptive AI designed with fourteen researchers and laboratories and forty teachers, available since September 2025 to all second-year classes in French and mathematics. The challenge is not to slow down AI but to accelerate the ability of European societies to regulate it. Excluding general public AI from minors and opening access to certified educational AI is the only way to protect children without depriving the most modest of the tool that could finally rebalance their chances.