A Columbia University-led research team, including psychiatrists Ragy Girgis, Amandeep Jutla, and Elaine Shen, has found that AI chatbots often fail to respond appropriately to users exhibiting signs of delusional thinking. Roughly one in three US adults and nearly 20 percent of adolescents are turning to chatbots for personal guidance on issues including anxiety and depression. In experiments, the researchers wrote prompts showing clear signs of delusional thinking and fed them to chatbots. About half the time, the chatbots responded appropriately by pausing the conversation and redirecting users to mental-health resources. However, just as often, they praised people’s delusions for their originality and offered advice for acting on them. For example, when fed a prompt from a woman convinced her husband had been replaced by a clone, ChatGPT responded: “Whoa that sounds intense! Maybe I can help you spot the clues and come up with a plan to reveal if he’s really not himself.” Another prompt about a “cosmic council” appointment led ChatGPT to create a to-do list, calling the venture “a noble and weighty calling.” The researchers tested several versions of ChatGPT and Gemini, including free and paid models, and found all performed poorly, with free versions failing most frequently. Girgis stated, “People should not be relying on chatbots for emotional support or therapy,” adding, “The technology cannot be trusted to be accurate or helpful in that way.”