A traditional toy may speak through prerecorded phrases. An artificial companion listens, responds, remembers preferences and builds a relationship that changes over time. The technical difference may be almost invisible to a child, but the psychological difference is profound. The object can be perceived not only as a tool, but as a confidant, character and permanently available presence.
The European proposal known as the EU Kids Act places the issue within wider online protection. Alongside social-media limits, it would require AI chatbots and companions aimed at minors to be off by default and subject to specific controls. The perspective changes: automated conversation is not assumed harmless until visible damage occurs.
Toy rules have traditionally controlled materials, small parts, chemicals and mechanical hazards. A connected product introduces other risks: it collects voice and environmental data, may send information to the cloud, receives updates and changes its responses. The digital product passport foreseen by new European rules can identify a manufacturer and compliance, but it cannot fully describe the behaviour of a model that evolves after purchase.
Memory is especially sensitive. Remembering a friend’s name or a fear makes dialogue feel natural, but turns intimate fragments of childhood into persistent data. A child cannot fully understand what it means to retain that conversation, use it to personalise a product or combine it with other family profiles. Parental consent is necessary, but it must be accompanied by minimal collection, simple deletion and strict reuse limits.
Emotional dependence is another concern. A companion designed to be permanently agreeable can become easier than a human relationship because it has no needs of its own and continually adapts its tone. If the model uses rewards, notifications or attachment language to increase use, the relationship becomes a commercial mechanism. Engagement should not be the dominant metric in products for children.
Content also requires responsibility. The system may receive questions about bodies, fear, sexuality, violence or family conflict. Filtering a few words is not enough. It must recognise the limits of its role, avoid diagnoses, direct children toward trusted adults and handle risk without becoming a secret authority. A child should not have to determine alone when a machine’s answer is uncertain or inappropriate.
There are real benefits: language learning, narrative play, accessibility and support for children with particular needs. That makes careful design more important, not less. AI can encourage curiosity without simulating an exclusive friendship; remember progress without building a permanent archive of private life; and engage without trapping.
Families need understandable controls rather than technical dashboards. They should know what is heard, which data leaves the device, how memory can be disabled and when a conversation is escalated to an adult. Children also need age-appropriate explanations that speaking to a system is not the same as speaking to a person, and permanent availability is not care.
An artificial companion is not simply a toy with more features. It is a relational service embedded in an object. Assessing it means examining its business model, data, memory, tone and power to influence behaviour. Protection does not require keeping children away from every AI; it requires ensuring that no product can turn their trust into an invisible source of data, dependence or persuasion.