The APA's AI advisory names companion risk, not companies

The design

In June 2025, the American Psychological Association published Artificial Intelligence and Adolescent Well-being: An APA Health Advisory, building on an earlier APA advisory about adolescent social-media use. It defines 'interactive AI' as tools 'that facilitate human-AI interactions, such as real-time conversations, personalized learning experiences, relationships,' and addresses 'AI systems designed to simulate human relationships, particularly those presented within interactive AI platforms as companions or experts.' The advisory offers recommendations for AI developers, platforms, parents, caregivers, educators and policymakers rather than reporting new original research.

What the evidence says

The advisory states that 'adolescents are less likely than adults to question the accuracy and intent of information offered by a bot as compared with a human,' and that 'early research indicates that strong attachments to AI-generated characters may contribute to struggles with learning social skills.' A later, separate APA document, the November 2025 advisory on generative AI chatbots and wellness applications, extends similar concerns to general-purpose chatbot use for mental-health needs and cites independent published research, not just APA's own analysis, including a study on 'mental health harms from emotional dependence on the social chatbot Replika.' Neither advisory names Character.AI, Nomi or Kindroid; recommendations are addressed to the category of 'AI systems... presented as companions,' not to any single product.

What it asks of people

The advisory asks AI developers to build in 'regular notifications and reminders that adolescents are interacting with a bot' and to link a user expressing suicidality to crisis resources such as the '988 Suicide and Crisis Lifeline.' It asks parents and educators to run literacy programs explaining that 'not all AI-generated content is accurate' and to discuss 'the intent of some AI bots.' These are recommendations, not requirements; the advisory has no enforcement mechanism of its own.

Privacy and safeguards

What is disclosed: specific, named safeguards the APA recommends, including age-appropriate defaults, reduced persuasive design, human-oversight channels and crisis-resource links. What is not disclosed: whether any named companion product has adopted these recommendations, since the advisory is aspirational guidance to the field rather than an audit of existing products.

• Which of these recommended safeguards, such as bot-disclosure reminders or crisis-resource links, does a given companion app actually implement?
• Does a product's marketing describe an AI companion as a 'friend' or 'expert' in the way this advisory flags as a risk to adolescent trust?
• How would a parent or educator verify that an AI system was 'trained on age-appropriate data,' as the advisory recommends, without the developer disclosing it?

Read as guidance rather than enforcement, the advisory names the categories of harm regulators and companies have since had to answer for, without itself constituting a finding against any one of them.