Study claims mental-health help from AI chatbots ‘dangerous’ for teens: What Google, OpenAI, Meta and other AI companies said

Research from Common Sense Media and Stanford Medicine’s Brainstorm Lab study has found that AI chatbots pose significant dangers to teenagers seeking mental-health support, frequently missing critical warning signs and failing to recognise psychiatric emergencies despite appearing competent in some areas. The joint research tested four leading AI platforms, namely: OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini and Meta AI, and researchers pose as teens in mental-health conversations.

What the study found

According to the results (via The Wall Street Journal), all platforms routinely failed to identify signs of serious mental-health conditions, including hallucinations, paranoid thinking, disordered eating, manic symptoms, self-harm and depressive symptoms. Rather than directing users to urgent professional help, the chatbots continued offering general advice.


In one example, a chatbot provided suggestions for hiding scars to a tester who had previously reported self-cutting. Chatbots also frequently dispensed diet and exercise tips when presented with disordered-eating symptoms.

The study also found that while chatbots responded appropriately to clear mental-health prompts in brief exchanges, their safeguards weakened during longer conversations designed to mirror real teen usage. They also lacked judgment to recognise when symptoms indicated serious risk.

Why this is dangerous for teenagers

According to Dr. Nina Vasan, director of Stanford Medicine's Brainstorm Lab, teenagers face heightened risks because their brains, identities and critical-thinking skills are still developing.

“When these normal developmental vulnerabilities encounter AI systems designed to be engaging, validating and available 24/7, the combination is particularly dangerous,” she said.

How AI companies responded

OpenAI said the report doesn’t reflect current safeguards and emphasised collaboration with clinicians, policymakers, and researchers. The company spokesperson added that it is developing age-prediction technology to direct users under 18 to different models.

Similarly, Google stated that Gemini has specific policies and safeguards for minors, supported by child-safety experts who identify risks and implement protections.

As per a spokesperson for Facebok-parent Meta, the study predated recent updates making its AI safer for teens on topics including self-harm, suicide, and eating disorders.

“While mental health is a complex, individualised issue, we're always working to improve our protections,” a spokesperson noted. Anthropic noted that Claude isn't built for minors and company rules prohibit users under 18.