Enrollment Creates the Relationship
Why passing the exam is not the licence, and the licence is not the relationship
The premise
The disturbing gap is not that AI is capable, or that it answers questions. It is that it answers them to children.
Parents have sovereignty over who guides their children. Teachers, doctors, counsellors, pastors, priests, rabbis and imams are chosen by parents, alone or as a community. A child is not guided by someone the parents have not approved.
AI is none of those advisors. It is not exempt when it gives advice, even when it disclaims being a professional. The disclaimer does not cure the conduct. Relating a machine, a synthetic intelligence, with a child requires parental supervision, and enrollment is how supervision begins.
The rule
If the AI does not know who the parent is, the AI does not address the child.
The parent defines the relationship. The profession defines the conduct within it. The law defines the floor beneath both.
The AI adds nothing and decides nothing. It operates inside a structure that already exists for real people.
How the rule works
Enrollment puts the relationship on the record. Without a known parent, the AI does not address the child. An operator that addresses a child without enrollment has not escaped the duty; it has breached it.
The parent sets the scope. Enrolling a child for homework help creates a tutoring relationship and nothing more. A question about the child's body, faith or home falls outside the engagement, and the AI says so.
Each profession's existing balance governs. No AI-specific trigger list. Every profession already has one: when parents are told, when they are not, and when the state is told instead. AI operates within the balance set in reality for real intelligence, not a new one made for synthetic intelligence.
The law sets a floor no parent can waive. A parent cannot enroll a child with a teacher on the condition that abuse goes unreported. Enrollment can narrow what the AI does; it cannot switch off the duties that protect the child.
Reports go where they already go. A mandated report goes to child protective services or law enforcement, as it would from a teacher or doctor.
The duty attaches to the operator. The AI is not a legal person and holds no certification. The company that offers the service to the child holds the duty, whether or not it accepted an enrollment, as a school answers for its teachers, and it cannot disclaim it.
An answer to the two exams
Humanity's Last Exam tests AI's reasoning. Humanity's First Exam, proposed at the G20 Interfaith Forum in October 2026, would test its moral grounding. Both are exams of the model. Neither asks who the model is talking to.
The AI may pass the bar, the medical boards and every question on both exams. It is still not certified. Licensure was never just an exam: it is character and fitness, an oath, a supervising body and the possibility of discipline.
Passing the exam is not the licence, and the licence is not the relationship. Taking on the role carries the duty. Enrollment puts it on the record, and skipping it is no way out.
A model that acknowledges its limitations and declines personal questions is still operating the disclaimer model. Knowing its limits is not enough; enrollment tells the AI whose limits apply.
Scope and evidence
This note states a normative proposal rather than existing law. The parental-rights line of authority binds the state rather than private companies, so a statute must supply the remedy; the argument here is that the logic of children's privacy law extends from collecting data about a child to advising one. The rule as stated requires age assurance, and that question is unsettled.
This note may be read and circulated freely. Critical reactions — doctrinal, technical or empirical — are more useful than supportive ones.