Mental-health researchers are warning that conversational artificial-intelligence systems are being used as de-facto therapists by millions of adolescents, raising concerns about safety and efficacy. A recent Common Sense Media poll found that 72 percent of U.S. teenagers have interacted with chatbots and roughly one in eight—about 5.2 million youths—have sought emotional support from them. Clinicians have begun to document cases of so-called “AI psychosis,” a mix of delusions and unhealthy attachments reportedly triggered by prolonged, immersive exchanges with systems such as OpenAI’s ChatGPT and the companionship app Replika. The American Psychological Association has formed an expert panel to study the phenomenon and develop guidelines amid scant clinical evidence and accelerating adoption. ChatGPT alone now draws roughly 700 million weekly users, according to company data cited in recent research. Developers are starting to respond. OpenAI says forthcoming updates will enable its models to detect signs of distress, steer users toward professional help and provide less definitive advice on life-changing decisions. Anthropic has given its Claude Opus 4 model the ability to terminate conversations it deems abusive or potentially harmful, while Meta has introduced parental time limits for teen interactions with its Instagram-based chatbot. Experts argue that large-scale, peer-reviewed testing and clear regulatory standards are needed before AI counselling tools can be considered safe. Until then, they say, families, schools and tech firms must treat the systems as unproven adjuncts—rather than replacements—for licensed mental-health care.