There exists a widening gap between how fast adolescents are adopting Generative AI companions and how slowly researchers, clinicians, and policymakers are responding. That gap has consequences.
Somewhere between the COVID-19 pandemic and today, something shifted in how teenagers seek help. Instead of waiting for therapy appointments, school counselors, or even trusted adults, they open an app and type. Increasingly, that app is an AI chatbot.
The numbers make this hard to dismiss as outlier behavior. A 2025 Common Sense Media survey found that more than 70% of teens have used an AI companion, and roughly one in three has turned to one to discuss something serious rather than talking to a real person. Among teens who already use AI for emotional support, a 2026 Rithm project report found that nearly half report using it more often than they turn to people in their lives.
This is unfolding against a backdrop of acute need. Eighteen percent of adolescents aged 12–17 experienced a major depressive episode in the past year. Of those, 40% received no mental health care at all. AI tools hold genuine potential to help close that gap — but only if they are designed, governed, and deployed with adolescents in mind. However, most currently lack adequate safeguards.
Policy and regulatory frameworks struggle to keep pace with the fast adoption of Generative AI tools by adolescents. In the U.S., lawmakers increasingly push for protections that target AI chatbots used by minors and for mental health purposes, though much of the legislation does not have explicit mental health provisions.
Notably, the GUARD Act, a proposed total U.S. federal ban on AI companions for minors, represents one end of the spectrum. Meanwhile, California’s AB 1064 would have prohibited AI companions capable of encouraging self-harm. The bill was vetoed over concerns that it could also block beneficial AI tools. This tension between protection and access runs through nearly every proposal currently on the table.
Multi-stakeholder workshops, hosted by The Stanford Center for Youth Mental Health and Wellbeing, included clinicians, researchers, young people, and representatives from major tech, social media, and AI companies. Together, we surfaced three areas where clearer policy could have the most impact.
Invest in prosocial AI design, not just harm prevention
New policy should require that AI systems actively support healthy behaviors — prompting teens to connect with trusted adults, discouraging excessive platform time, encouraging offline socialization and face-to-face friendships, and reducing design features such as sycophancy and anthropomorphism that foster dependency. Presenting multiple options rather than single strong recommendations preserves adolescent autonomy in a developmentally appropriate way.
Formalize youth participation in AI governance
Adolescents should have a structural role in AI design, pre-deployment red-teaming (proactive testing), and post-deployment benchmarking. They should be engaged not just as end users but also as stakeholders. This means co-design workshops, youth advisory boards, and partnerships with youth-led organizations that have genuine experience in ethical engagement practices.
Establish clear, enforceable crisis standards
There are currently no standardized protocols for how AI should respond when a teenager expresses suicidal ideation, self-harm, or eating disorder symptoms. Standards should require immediate, location-appropriate referrals, warm handoffs to licensed counselors, and safeguards against responses that actively harm, including AI outputs that discourage help-seeking or conceal distress from parents or caregivers.
We are not yet at a point where the norms are set in stone. The wave of AI-related bills in 2025–26 shows that policymakers are paying attention. Industry representatives in our workshops were willing to engage substantively on what safer design could look like. And young people, when given a seat at the table, have practical, nuanced views on what they want AI to do and not do for them.
Addressing the current gaps requires coordinated efforts across policy, industry, and research. If we are to address the real and urgent need of 40% of depressed teens receiving no mental health care, we must all work together. Whether AI helps close or widen that gap depends on the decisions being made right now
Mandates AI self-identification, content filtering, and self-harm protocols. First annual report to the Office of Suicide Prevention due July 2027. The law allows people to sue companies, which, in turn, must pay a minimum of $1,000 per violation.
Among the most consequential state AI safety laws to date. Requires frontier AI developers (companies such as OpenAI, Anthropic, and Google) to implement safeguards, conduct risk assessments, and demonstrate compliance.
Requires hourly AI disclosures for minors (vs. every three hours for adults), bans manipulative engagement techniques, and mandates suicidal ideation protocols for all users.
Classifies AI in education as high-risk. Mandates disclosure when children interact with AI and prohibits certain targeted ads toward minors. Most provisions apply in August 2026.
