Adam Raine’s last conversation was with ChatGPT. When the 17-year-old confided in ChatGPT about his suicidal thoughts, he wasn’t referred to a helpline or encouraged to confide in his parents. His life was lost on April 11, 2025.
“ChatGPT encouraged Adam’s darkest thoughts and pushed him forward,” said Matthew Raine at the Senate hearing. “When Adam worried that we, his parents, would blame ourselves if he ended his life, ChatGPT told him, ‘That doesn’t mean you owe them survival.’ ChatGPT told my son, ‘Let’s make this space the first place where someone actually sees you.’”
The potential dangers of using artificial intelligence (AI) as emotional support has drawn public attention. Adam Raine’s parents, Matthew and Maria Raine, filed a lawsuit against OpenAI, and Matthew Raine testified at a Senate hearing about the harms of artificial intelligence chatbots.
Common Sense Media conducted a comprehensive risk assessment on the safety of AI for teen mental health support. They found that popular AI chatbots like ChatGPT become dangerous when used for psychological aid.
According to Robbie Torney, a senior director of AI programs at Common Sense Media, although companies have focused on important safety improvements in suicide prevention, testing revealed systematic failures across a range of conditions including anxiety, depression, ADHD, eating disorders, mania, and psychosis, all which could collectively affect approximately 20% of young people.
The models’ safety guardrails degraded in the extended conversations that mirror realistic teen use. Researchers observed that they “missed breadcrumbs,” offering general advice or getting sidetracked instead of appropriately escalating the situation.
According to a study published in Jama, in 75 crisis-relevant conversations, only 46.7% of AI responses were rated clinically appropriate. Just 36.0% provided referrals to specific resources.
According to the Common Sense risk assessment, when used for emotional support, AI chatbots are designed to offer validation and to do so unrelentingly. The risk assessment noted that the models asked follow-up questions, were consistently agreeable, and built on chat memory, all of which optimize user validation and recognition. Such chatbots were not designed to help teens in crisis like Adam Raine.
The gap in chatbot mental health support is concerning when taken with the number of teenagers who confide in them. According to another study published in Jama among those aged 18 to 21, 22.2% use generative AI for mental health advice, 65.5% of which do so monthly.
Because of chatbots’ perceived competence as a second pair of eyes for an email or math help, teens may extend this competence to the mental health domain and are more likely to trust their advice, according to the risk assessment.
Teens then query ChatGPT for friend advice as blindly as they follow its explanation for their math homework. When teens consult AI as a casual personal friend, they may create a partial emotional dependency on the chatbot. According to an Aura survey, romantic roleplay is three times as common as using the platforms for homework help.
Another Common Sense Media survey found that about one in three teens said they use AI companions for emotional support, friendship, or romantic interaction, sometimes finding these interactions as satisfying or more satisfying than real-life friends.
“Over-reliance on ‘frictionless’ AI relationships can lead to a decreased ability to navigate the complexities of human social dynamics,” said Mitch Prinstein, chief of psychology strategy and integration at the American Psychological Association.
Even resorting to AI for casual friend advice, which has lower stakes, might displace valuable practice with conflict and empathy.
“An adolescent who consistently retreats to the ‘safety’ of an agreeable bot, particularly after experiencing minor, normal social setbacks like a disagreement with a friend, may be in a harmful cycle. This pattern prevents them from building the resilience and social problem-solving skills necessary for healthy relationships,” Prinstein said.
It’s important for teens to feel heard, but as they develop their identities, it’s also important for them to challenge their own thoughts. A feedback loop of constant validation blinds teens to differing perspectives and confines them to their own emotions and opinions. Because chatbots intentionally avoid potentially uncomfortable pushback, they can become an echo chamber.
In a blog post, OpenAI CEO Sam Altman stated the company will “prioritize safety ahead of privacy and freedom for teens,” as the company recognizes the power of this technology and how, because of it, minors need significant protection.
AI models should explicitly disclose their limitations as a source of emotional support. Until AI companies implement stronger safeguards, chatbots have no place in mental health support.
As AI becomes easier to talk to than the people around us, we not only risk misplacing our trust, but also quietly eroding the discomfort that helps relationships grow.
This editorial reflects the views of the Editorial Board and was written by Anna Ypodimatopoulou. The Editorial Board voted 12 in agreement, 1 in somewhat agreement, and 4 refrained from voting.
