🔎︎
A report published by STAT News on April 13, 2026, examines a growing movement among hospitals to deploy their own patient-facing AI chatbots rather than ceding early clinical conversations to commercial platforms like ChatGPT, which now handles more than 40 million health queries daily. Hartford HealthCare this week launched PatientGPT, built by K Health, for tens of thousands of established patients in Connecticut. Two other systems, Sutter Health and Reid Health, are piloting Emmie, the chatbot developed by electronic health record vendor Epic.
These hospital-branded tools are designed to pull from patients' existing medical records and route them to appropriate care within the system, whether that means urgent care, an emergency department, or a telehealth visit. Mount Sinai Chief AI Officer Girish Nadkarni described this as "structurally safer" than commercial chatbots that lack a clinical backup pathway. However, health systems also assume greater liability than consumer-facing AI companies if something goes wrong.
Safety testing remains an early and unfinished process. A red-teaming analysis of PatientGPT found an 8.5% failure rate in high-risk scenarios, including under-triaging urgent symptoms and hallucinating clinical context. A subsequent 400-patient pilot study revealed no apparent safety issues, though researchers questioned whether the sample size was sufficient to capture the range of real-world risks.
Epic's Emmie may hold a structural advantage because it operates through the widely used MyChart portal. Meanwhile, companies like Amazon and General Medicine are also using chatbots as entry points to care. The competitive landscape is prompting health systems to move quickly, even as clinical evidence supporting the safety and efficacy of these tools remains limited.
