Children and AI Companions: The Policy Fight Nobody Was Ready For

What the data shows

Independent surveys and app-store telemetry converge on the same picture: a large minority of teenagers in high-income countries use at least one AI companion app on a weekly basis, and a smaller cohort uses one daily for extended sessions. The use patterns look less like social media and more like intimate communication: private, emotionally significant, and disclosed to few adults.

The clinical evidence is early and mixed. Some studies report reduced loneliness among heavy users; others report increased social withdrawal and disordered attachment. Neither literature is mature enough to support strong policy claims — but the underlying use is spreading faster than the evidence base.

Why existing frameworks do not fit

Child safety online has historically focused on content moderation, contact restrictions, and advertising limits. AI companions blur every one of those categories: the content is generated on demand, the contact is with a system rather than a person, and the monetization is often subscription rather than advertising.

The result is a policy vacuum. Regulators trained on the last generation of platform harms find the tooling — takedown, moderation, ad restrictions — does not map cleanly onto the new pattern.

• Age verification remains unreliable, especially for teenagers motivated to work around it.
• Content moderation on generated conversation is fundamentally different from moderation on static posts.
• The intimate nature of the interactions raises specific concerns about therapeutic claims and unlicensed practice.

The interventions on the table

Three regulatory approaches are competing. The first is disclosure-based: require companion products to state clearly that they are not human, not therapists, and not appropriate for certain age groups. The second is design-based: mandate specific product features — session limits, escalation to crisis resources, refusal of certain topics — that apply regardless of business model. The third is access-based: restrict certain product categories to specific age groups, with real enforcement.

Each approach has trade-offs. Disclosure is the easiest to pass and the weakest in effect. Design mandates raise the specter of frozen product architectures. Access restrictions collide with the same enforcement problems that have plagued age gates for the last two decades.

“The children using these products every day are not waiting for the policy debate to conclude. The debate is happening about them, not with them.”

The industry response

The most responsible operators have moved ahead of regulation with age-tiered product designs, escalation to human support for crisis signals, and voluntary limits on certain interaction patterns. The least responsible have moved in the opposite direction, treating young users as a growth engine.

The gap between the two ends of the industry is the single most important variable in what the regulatory outcome will look like. If the responsible cluster establishes a de facto standard, regulation may codify it. If the least responsible operators define the public perception of the category, regulation is likely to be blunter and more restrictive.

What to watch

Watch for the first jurisdictions to move from consultation to statute. Watch for coordinated action across pediatric societies, education systems, and child-welfare agencies — historically the coalitions that shape how children's technology gets regulated. And watch the class-action bar: private litigation may reshape the category faster than any statute.