In a May 14 STAT First Opinion essay, addiction medicine physician Steve D. Klein argues that patient-facing conversational AI “agents” may be particularly risky in addiction treatment, even if they sound caring or empathetic. Klein describes being intrigued by a product that could text patients between visits, summarize check-ins into metrics, and help clinicians monitor day-to-day changes — a capability that could be valuable in recovery care. But his concerns grew when the company touted that patients form emotional relationships with the AI agent. Klein warns that recovery depends on rebuilding real human connection and community supports (sponsors, therapists, home groups, loved ones), and that encouraging emotional attachment to an algorithm could divert limited “emotional equity” away from relationships that actually sustain long-term recovery.
He also challenges the idea that an AI can truly offer bedside manner, emphasizing that large language models can simulate empathy and concern without accountability or genuine investment in a patient’s well-being. Klein flags additional ethical concerns when these agents are labeled “doctor” or “M.D.,” calling it misleading. His bottom line: AI can be transformative for provider-facing tasks like documentation, but “unfiltered” patient-agent interaction is a dangerous fit for sensitive behavioral health populations.
Read More: https://www.statnews.com/2026/05/14/addiction-medicine-ai-agent-empathy-caring-misguided/