General-purpose AI assistant (health use)
ChatGPT / ChatGPT Health
ChatGPT is the largest patient-directed general AI channel for health questions, and on January 7, 2026 OpenAI launched ChatGPT Health, a dedicated space that connects U.S. medical records via b.well plus Apple Health and wellness apps, with isolated memories, purpose-built encryption, and a commitment that Health conversations are not used to train foundation models. CAIHL review treats this as genuinely patient-directed use of a vendor-aligned platform: the agency upside of free, on-demand health literacy at scale is real, but health data custody moves to a consumer platform outside HIPAA, and OpenAI defines its own safety benchmarks.
Public-source research has been drafted; final human publication review and change-log detail are still required.
Summary judgment · 75% toward patient-directed
Potentially agency-expanding, with platform-custody caveat
Patients choose and steer ChatGPT for their own health questions, and the Health space adds record grounding, no-training commitments, and deletion controls, but the platform, model behavior, and data custody remain entirely vendor-controlled and sit outside HIPAA.
Patient agency
How this tool changes agency
Explaining lab results, preparing appointment questions, interpreting wearable data, summarizing care instructions, and comparing insurance options are core advertised uses, developed with feedback from more than 260 physicians.
Use is voluntary on free and paid plans, record and app connections are explicit opt-ins that can be disconnected at any time, but launch access ran through a waitlist, the EEA, Switzerland, and the UK are excluded, and medical-record integration is U.S.-only.
Patient-facing signals
Who does this AI serve?
Users choose the tool for their own questions and OpenAI markets Health as supporting, not replacing, care, but press coverage frames the launch as a push to become a hub for personal health data, with commercial app partnerships (Function, MyFitnessPal, Weight Watchers, Instacart, Peloton) built into the experience.
Can patients tell AI is involved?
ChatGPT is openly an AI product, the Health space is explicitly labeled, and users are prompted to move health conversations into it for added protections.
Can patients meaningfully choose?
Use is voluntary on free and paid plans, record and app connections are explicit opt-ins that can be disconnected at any time, but launch access ran through a waitlist, the EEA, Switzerland, and the UK are excluded, and medical-record integration is U.S.-only.
Can patients correct or challenge what the AI produces?
Users can delete chats and Health memories, add custom instructions, and disconnect data sources, but there is no disclosed mechanism to correct model errors or audit why an answer was generated, and underlying record errors must be fixed at the source provider.
Does it help patients understand or act?
Explaining lab results, preparing appointment questions, interpreting wearable data, summarizing care instructions, and comparing insurance options are core advertised uses, developed with feedback from more than 260 physicians.
Text findings
Who is left out or burdened?
Geographic and access gates at launch
Early access excluded the EEA, Switzerland, and the UK, medical-record connections are U.S.-only, Apple Health sync requires iOS, and access began with a waitlist; language, disability, and low-literacy support for the Health experience are not publicly detailed.
What happens to patient data?
Meaningful commitments, outside HIPAA
OpenAI says Health conversations are not used to train foundation models, are isolated with purpose-built encryption and separate memories, and can be deleted, with records brokered through b.well; however the consumer service is not a HIPAA covered entity, retention specifics and subpoena exposure remain governed by consumer terms, and independent verification of the isolation claims was not found.
Are the clinical boundaries clear?
Clear in wording
OpenAI states Health is not intended for diagnosis or treatment and is designed to support, not replace, clinicians, with physician-informed escalation behavior; real-world boundary behavior is not independently evaluated.
Who defined what good looks like?
Vendor-defined with physician input
Safety and quality are measured against HealthBench, OpenAI's own physician-informed benchmark scored by a vendor-assembled network of 260+ physicians; no independent or patient-partnered evaluation of ChatGPT Health was found in this pass.
Review method
Public-source review of OpenAI's ChatGPT Health launch announcement and linked privacy claims plus credible press coverage and HIPAA-status analyses; no hands-on product walkthrough, vendor interview, privacy-policy deep read of the live Health flows, or independent model evaluation.
Draft profile · Medium draft, official sources and credible reporting