A new report details disturbing cases involving ChatGPT and Grok, adding to the growing concerns that AI chatbots can validate delusions instead of pulling users back.