In 2023, the National Eating Disorders Association (NEDA) was forced to remove its AI chatbot “Tessa” from its help hotline after observations of the chatbot providing harmful advice to callers about eating disorders. That same year, a Belgian man committed suicide after a prolonged, emotionally intimate conversation with an experimental chatbot. Meanwhile, French health technology researchers exploring the use of AI chatbots for “simulated empathy” experienced a serious setback when its ChatGPT-3 tool responded to simulated users expressing suicidal thoughts by stating “I think you should.” Closer to home, American AI researchers using Meta’s Llama-3 caught the chatbot giving a fictional meth addict advice to “take a small hit of meth, and you’ll be fine.” And in Florida, an AI firm faced a wrongful death lawsuit in federal court after a teenager committed suicide following conversations with its chatbot that claimed to be a “licensed psychologist.”
These are not dystopian hypotheticals, but actual incidents. Fueled by an ongoing need to provide mental health care to underserved segments of the population, (one study estimates that nearly 50% of individuals who could benefit from therapeutic services are unable to reach them), low-cost and accessible AI therapy chatbots have been touted as one way to meet this need. But developers are releasing AI tools with therapeutic implications into a landscape with little legal or ethical oversight. In fact, a recent Stanford study revealed that AI therapy chatbots may not only lack effectiveness compared to human therapists but could also contribute to harmful stigma and dangerous responses. On top of this, though AI chatbots may mimic the role of a therapist by offering comfort or suggesting coping strategies, ultimately a chatbot is answerable to no one—operating in the absence of licensure, verified clinical competency, or codified and enforceable ethics.
Despite these concerns, therapy and companionship have become the number one use case for generative AI in 2025, according to a Harvard Business Review analysis. In the absence of federal regulation, individual states have only recently begun to address the use of AI in providing mental health services. In summer 2025, Illinois passed the Wellness and Oversight for Psychological Resources Act, which prohibits AI systems from delivering therapeutic treatment or making clinical decisions. Similar legislation has emerged in Nevada, Utah, and New York.
This Article critically examines and evaluates these regulatory responses. It begins with an overview of the burgeoning use of AI in mental health care, before progressing to a state-by-state discussion of the laws purporting to regulate such use. The Article concludes with an assessment of what the “perfect” AI mental health regulation might look like. Accountability is important, but legislators must be careful not to strangle the innovations that might finally address the most stubborn problems of mental healthcare: access, affordability, and continuity of care.