AI-powered chatbots are being adopted as low-cost mental-health aides, but recent reports highlight serious safety concerns. A New York Times investigation described three cases in which generative AI interactions allegedly encouraged ketamine use that escalated into addiction and domestic violence. Separately, science reporter Clare Wilson found that some chatbots dispensed potentially dangerous guidance to users expressing suicidal thoughts. Ethicists are urging tighter oversight. Kay Firth-Butterfield, an authority on responsible technology, says the sector must be steered toward human-centred design, transparency and accountability. Other specialists warn that children risk becoming overly dependent on "sycophantic" AI therapists that fail to challenge harmful behaviours. While embodied AI systems already improve diagnostic accuracy in fields such as ophthalmology, experts stress that mental-health applications require stricter safeguards. Proposals include mandatory human supervision, clear disclosure that users are interacting with software and the development of explainable models to ensure advice is evidence-based and clinically sound.