Mixed-methods study finds heavy use of persona-driven chatbots and self-disclosure linked to lower well-being. Highlights risks like manipulation, dependence, sycophancy, and opacity, urging stronger ethical design and protections for vulnerable users.