Children in the Behavioural Machine: A Survey of Behavioural AI in the Environments of Childhood and the Case for Child-First AI Governance

Systems that predict, rank, recommend, personalise and influence behaviour now structure much of the daily experience of children in the places where they learn, play, communicate and spend. This paper examines how such systems affect children differently from adults, and asks whether the technologies actually deployed in schools, games, social platforms, companion chatbots and commercial services, together with their operators' published policies, provide age-appropriate safeguards, meaningful transparency and accountability where behavioural data is used to make or shape consequential decisions. Drawing on a documentary survey of thirty deployed systems and enforcement records across the United States, Canada, the United Kingdom, the European Union, Australia and Asia, and on peer-reviewed research in developmental science, education, law and digital rights, the paper identifies a consistent pattern. Child-specific protections, where they exist at all, cluster around advertising and parental consent; they rarely reach the ranking, prediction and classification logic that determines what a child sees, how long she stays, what label is attached to her, and who is told. Transparency runs to parents and institutions rather than to children. Accountability arrives after harm, through litigation discovery, regulator action and adversarial civil-society audits, rather than being designed into deployment. The paper argues that these gaps are not incidental but structural: they follow from a business logic that treats engagement and prediction as the product, from consent architectures ill-suited to children's evolving capacities, and from the durability of algorithmic classification in institutions that children cannot leave. It closes with eight policy recommendations intended to place the best interests of the child at the centre of AI governance, including a prohibition on engagement optimisation for minors, a right of contestation for consequential classification, mandatory expiry of behavioural labels, and child-readable disclosure of inference.