AI Policy Wiki
Dashboard
⚲
AI Policy Wiki › Source Summaries

Artificial Intelligence and Adolescent Well-being: An APA Health Advisory (June 2025)

medium confidence · updated 2026-08-10

Primary text of the American Psychological Association's June 3, 2025 health advisory on AI and adolescent well-being — seven background considerations, nine recommendation clusters addressed to developers, policymakers, educators and caregivers, and a 13-member expert advisory panel.

Contents

A health advisory published June 3, 2025 by the American Psychological Association (APA), addressing the association of artificial intelligence with adolescent psychological development and setting out recommendations for developers, policymakers, educators, parents and caregivers, and adolescents themselves. It was produced by a 13-member expert advisory panel with APA staff leads Mitchell J. Prinstein (chief of psychology), Corbin Evans (deputy chief of advocacy for science and technology), and Ludmila Nunes (senior director for scientific knowledge and expertise), and carries a 69-item reference list (Source: apa.org).

The advisory is the second in a sequence: it builds on APA's earlier health advisory on adolescent social-media use, and precedes the separate November 2025 advisory on generative AI chatbots and wellness applications for mental health, summarized at APA Health Advisories on AI and Adolescent / Mental-Health Well-being (2025).

Scope and definitions

The advisory defines adolescence as "generally considered to be between ages 10 and 25" — a wider band than the legal-minority definitions used in most AI statutes. It distinguishes two categories of system:

  • Generative AI — applications that generate humanlike text, photorealistic images, lifelike audio, and realistic video.
  • Interactive AI — platforms or tools facilitating human-AI interaction, including real-time conversation, personalized learning, relationships, and tailored content recommendation.

It treats a third category separately: AI used to automate consequential decisions about young people, naming school admissions, medical diagnosis, and automated grading.

Background considerations

The advisory states seven premises on which its recommendations rest:

  1. Effects on adolescent development are "nuanced and complex"; AI is "not all 'good' or 'bad'." Assessment must account for the specific application, design features, training-data uses, and context of use.
  2. Adolescence is a long developmental period, and age "is not a foolproof marker for maturity or psychological competence" — two adolescents of the same age are unlikely to be at the same developmental level.
  3. Adolescence is a period of brain development in which adolescents experience "greater developmental changes than any other period in life besides infancy."
  4. Individual differences — temperament, neurodiversity, exposure to stress or violence, social isolation, traumatic experience, mental health, age, and socioeconomic or structural disadvantage — mediate response to the same content.
  5. Bias and discrimination are "often present in AI output" because of unrepresentative training data, lack of diversity in product testing and development, and the fact that programming "has been designed by adult humans."
  6. Adult use of AI, by parents and teachers, influences adolescents' own attitudes and practices; adults are asked to model healthy behavior and critical thinking.
  7. Youth safety should be addressed early in AI's evolution so as "not to repeat the same harmful mistakes that were made with social media." The advisory gives two reasons AI differs: adolescents may be unaware they are using AI or AI-assisted technology at all, and AI can produce inaccurate information in ways that make users believe it is true.

Recommendations

The advisory organizes its recommendations into nine clusters, each addressed to a mix of developers, regulators, educators, and caregivers.

Healthy boundaries with simulated human relationships. Systems designed to simulate human relationships — companions, or chatbots presenting as experts or as providers of social or mental-health support — must incorporate safeguards. The advisory gives two grounds. First, adolescents are "less likely than adults to question the accuracy and intent of information offered by a bot as compared with a human," may struggle to distinguish simulated empathy from genuine human understanding, and may not perceive a system's persuasive intent or bias; the result is "heightened trust in, and susceptibility to, influence from AI-generated characters, particularly those that present themselves as friends or mentors." Second, relationships with AI entities "may displace or interfere with the development of healthy real-world relationships." The advisory states that early research indicates strong attachments to AI-generated characters may contribute to difficulty learning social skills and developing emotional connections, while acknowledging that such technologies may offer companionship to youth who have difficulty forming relationships with humans of similar interests or identities. Specific measures: developer features preventing exploitation, manipulation, and erosion of real-world relationships including with parents and caregivers, such as regular notifications that the user is interacting with a bot and prompts toward human interaction where an adolescent indicates significant challenges such as suicidality or threats of abuse; regulatory measures ensuring such systems protect adolescent mental and emotional health; and literacy programs by educators and parents covering the inaccuracy of some AI content, the intent behind some bots, and indicators of misinformation or marketing goals underlying human-bot interaction.

AI for adults should differ from AI for adolescents. Systems "designed for, or foreseeably accessed by" adolescents must be built with explicit consideration of adolescent competencies and vulnerabilities, avoiding exploitation of heightened social sensitivity and underdeveloped impulse control. Five design measures are named: age-appropriate defaults (privacy settings, interaction limits, and content set to the most protective levels by default); transparency and explainability, including opt-in/opt-out for certain features and regular reminders that the user is interacting with non-human technology; reduced persuasive design, minimizing or eliminating gamification, personalized responses, and manipulative notifications in systems intended for or easily accessible by youth; human oversight and support, with the 988 Suicide and Crisis Lifeline named as an example of the resource a user expressing suicidality should be linked to; and rigorous testing with diverse groups of young users before widespread release, which the advisory suggests be accomplished by seating scientists, youth, ethicists, health professionals and other stakeholders charged with adolescent protection on AI-company advisory boards.

Uses of AI that promote healthy development. The advisory identifies benefits: AI can assist in brainstorming, creating, organizing, summarizing and synthesizing information, and can support cognitive development through advanced questioning that stimulates critical thinking, scaffolding that provides step-by-step guidance, and adaptive learning delivering personalized feedback. These benefits are conditioned on teachers having the skills to use AI appropriately and on AI not overriding "processes necessary for students to learn." The countervailing risks named are inaccurate AI-generated summaries, over-dependence impeding development of students' own knowledge and skills, and AI missing subtle verbal and nonverbal cues. The advisory recommends students actively question and challenge AI-generated content and use AI to supplement rather than replace existing strategies, engaging in active learning.

Limits on harmful and inaccurate content. Citing APA's video-content recommendations, the advisory states that exposure to harmful content is associated with increased risk of anxiety, depression and other mental-health problems, and that adolescents exposed to violence and graphic content may become desensitized or traumatized. It also states that repeated exposure to misinformation makes it more likely to be believed and to spread, and may hinder analytic thinking. Measures: robust developer protections against youth exposure to content that is age-inappropriate, dangerous, illegal, biased or discriminatory, or liable to trigger similar behavior in vulnerable youth; user reporting and feedback systems allowing adolescents and caregivers to customize content restrictions; educational resources for recognizing and avoiding harmful content; and collaboration with mental-health professionals, educators and psychologists on content-filtering mechanisms.

Accuracy of health information. The advisory states that young people often seek health information online and that misinformation or incomplete information can lead to harmful behaviors, misdiagnoses, and delayed or incorrect treatment. Systems providing health information to youth should either ensure accuracy and reliability or "provide explicit and repeated warnings that the information may not be scientifically accurate," with the advisory adding that sources purporting to offer empirically based information "vary significantly in quality and accuracy and should not be weighted equally in the training of AI models." It further recommends clear disclaimers that AI-generated information is not a substitute for professional health advice, diagnosis or treatment; resources and reminders directing adolescents to a human such as an educator, school counselor or pediatrician to verify information; and continual reminders from parents and educators that online and AI content may be inaccurate, persuasive by design, or harmful.

Data privacy. Systems collecting or processing adolescent data "must prioritize their privacy and well-being over commercial profit," maximizing transparency and user control. Platforms should limit use of adolescents' data for targeted advertising, for "personalized marketing that exploits their immature brain development," for sale to third parties, or for any purpose beyond that for which it was explicitly collected. Informed consent from users and caregivers is called essential. The advisory specifically extends the category of sensitive data to "biometric and neural information from emerging technologies," on the ground that such data "can provide insights into mental states and cognitive processes."

Protection of youth likenesses. Misuse of adolescents' images and voices can produce cyberhate, cyberbullying, and sexually abusive material including deepfakes and non-consensual explicit images, with the advisory associating these with increased risk of depression, anxiety, and suicide-related behaviors. It calls for stringent platform restrictions on use of youths' likenesses encompassing "both the input and output of content into AI platforms," with compliance-monitoring and enforcement mechanisms; instruction by parents, caregivers and educators on the dangers of posting images and on strategies for confronting disturbing, inappropriate or illegal images of peers or themselves; and school policies managing creation and proliferation of hateful AI-generated content.

Empowering parents and caregivers. The advisory notes caregivers "often have limited time or capacity" to learn about the age-appropriateness, safety, prevalence, and risks and benefits of these technologies. It calls for industry, policymakers, educators, psychologists and health professionals to collaborate on accessible resources that go "beyond simple ratings" to include detailed explanations of data-collection practices, algorithmic biases, and manipulative or addictive design elements, plus guidance on how to discuss AI with adolescents. It asks for customizable parental controls and interactive tutorials analogous to existing rating systems for movies, video games and music, regularly updated, and paired with default settings and parental tools for setting parameters and maintaining visibility into potentially harmful interactions.

Comprehensive AI literacy education. The advisory treats AI literacy as covering what AI is, how it works, its benefits and limitations, privacy concerns, and the risks of overreliance, with "a specific focus on algorithmic bias": how bias becomes embedded through skewed training data, flawed model design, or unrepresentative development and testing teams, and how it can produce incomplete or inaccurate information perpetuating "myths, untruths, and/or antiquated beliefs," or discriminatory information particularly regarding vulnerable groups. Responsibilities are divided three ways. Educators: integrate AI literacy into core curricula spanning computer science, social studies and ethics; provide teacher training on AI concepts, algorithmic bias and responsible use; offer hands-on experience with critical evaluation of outputs; and facilitate discussion of ethical implications. Policymakers: develop national- and state-level AI-literacy guidelines, allocate funding for AI-literacy resources and teacher training, "enact legislation that mandates age-appropriate AI literacy education in schools," and promote public-awareness campaigns. Technology developers: create accessible explanations of algorithms and data-collection practices, develop educational tools explaining algorithmic bias, collaborate with educators on age-appropriate curricula, incorporate bias detection and mitigation tools, and provide simple reporting mechanisms for suspected bias.

Research funding. The advisory calls for "significant and sustained investment" in five research areas: longitudinal studies tracking developmental trajectories of adolescents interacting with AI over time; designs enabling identification of causal relationships and long-term effects; diverse-population studies extending to younger children and marginalized and vulnerable populations; data accessibility and transparency, specifically "mechanisms for independent scientists to access relevant data, including data held by technology companies," covering algorithmic functions, content moderation and user-engagement metrics; and interdisciplinary collaboration among psychologists, neuroscientists, computer scientists, ethicists, educators, public-health experts, youth and caregivers.

Expert advisory panel

The 13 panel members were Mary Alvord (Alvord, Baker & Associates; Resilience Across Borders), Dawn Bounds (University of California, Irvine), Linda Charmaraman (Wellesley Centers for Women, Wellesley College), Sophia Choukas-Bradley (University of Pittsburgh), Dorothy Espelage (University of North Carolina at Chapel Hill), Ellen Fitzsimmons-Craft (Washington University in St. Louis), Joshua A. Goodman (Southern Oregon University), Jessica Hamilton (Rutgers University–New Brunswick), Mary Ann McCabe (APA Board of Directors member-at-large), Jacqueline Nesi (Alpert Medical School of Brown University), Grant J Rich (past president, Society for Media Psychology and Technology), Brendesha Tynes (University of Southern California), and S. Craig Watkins (University of Texas at Austin). Eight APA staff consultants are also listed.

Sources cited by the advisory

The reference list situates the advisory in the adolescent-development and media-effects literature rather than the AI-safety literature. Recurring anchors include Orben, Przybylski, Blakemore and Kievit on windows of developmental sensitivity to social media (Nature Communications, 2022); Crone and Konijn on media use and brain development (Nature Communications, 2018); Valkenburg and Peter's differential-susceptibility-to-media-effects model (2013); Fazio, Rand and Pennycook on repetition increasing perceived truth (2019); Thiel, Stroebel and Portnoff on generative machine learning and child sexual abuse material (Stanford, 2023); and the Hopelab, Common Sense Media and Harvard Center for Digital Thriving survey of teen and young-adult perspectives on generative AI (2025). Two policy instruments are cited directly: the White House Office of Science and Technology Policy Blueprint for an AI Bill of Rights (2022) and the 5Rights Foundation Children & AI Design Code. NIST Special Publication 1270, Towards a standard for identifying and managing bias in artificial intelligence, is cited in the AI-literacy cluster.

Relationships

Provenance

Verbatim primary text of the 11-page published PDF, retrieved 2026-08-10 and corroborated against the APA landing page and the June 2025 APA press release "APA calls for guardrails, education, to protect adolescent AI users" (Source: apa.org). Confidence is medium rather than high: the advisory's own text is unambiguous, but its empirical claims about AI specifically — as distinct from the social-media literature it draws on — are described by the APA itself as resting on research that "is still developing," and the advisory offers no quantitative findings of its own. Where the advisory characterizes evidence, it does so with the qualifier "early research indicates."

An earlier compilation of APA materials, Raw Sources/APA Health Advisory AI Adolescent Well-being 2025.md, merged this advisory with the November 2025 chatbots and wellness-apps advisory and states on its face that it is not a verbatim copy; it remains the source file for APA Health Advisories on AI and Adolescent / Mental-Health Well-being (2025), which covers the November advisory and APA's policy activity.

Categories: Source Summaries | us | framework | safety | ethics | nonprofit | youth
What links here (8)