OpenAI Opens ChatGPT Health to US Users

OpenAI Opens ChatGPT Health to US Users — and Medical Records Lose HIPAA Protection on the Way In

The January pilot failed for an unexpected reason: 70% of health questions happened outside the dedicated health tab. So OpenAI tore down the tab.

Jul 24, 2026Source: OpenAI

OpenAI began rolling out Health in ChatGPT to US users on Thursday at 10:00 a.m. Pacific, letting anyone logged in and over 18 connect Apple Health data and hospital records to the main chat interface. It is available on Free, Go, Plus and Pro, on web and iOS. There is no Android version yet, and it is not available in the EU.

Why the first attempt didn't stick

OpenAI piloted ChatGPT Health as a separate hub behind a waitlist on January 7. It was never withdrawn — it just sat there. The telling number: more than 70% of health-related conversations were happening outside the dedicated space, while people planned meals or searched recipes. The rebuild removes the hub and lets ChatGPT reach connected health data from any conversation.

The plumbing

The connectivity layer is b.well Connected Health, a Baltimore company whose network reaches 2.2 million US healthcare providers and 320 health plans. Supported systems include Epic, Oracle Health, One Medical, Kaiser Permanente and Function Health; records from unsupported providers can be uploaded manually. Wellness integrations — Apple Health, MyFitnessPal, Weight Watchers, Peloton, AllTrails, Instacart — carried over from January.

The privacy asterisk

OpenAI says health data lives in an additionally encrypted space with its own history and memories, is not used to train foundation models or target ads, and is deleted within 30 days of disconnection. What it cannot say is that the data stays protected: once records leave a health system, HIPAA no longer applies, and consumer accounts get no business associate agreement. EPIC senior counsel Sara Geoghegan called removing that protection "dangerous."

The safety work, and the lawsuit

OpenAI says 260+ physicians in 60 countries reviewed model outputs more than 600,000 times across 30 health domains. The launch landed a day after a former Florida pastor sued OpenAI in San Francisco Superior Court over ChatGPT health advice, asking a court to halt Health in ChatGPT pending independent safety audits. Roughly 300 million people a week already bring health questions to ChatGPT.

Why It Matters

The interesting finding here is not the product but the pilot data: people don't compartmentalize health questions, so a dedicated health mode was the wrong shape. The consequence is that medical records now flow into a general-purpose assistant governed by terms of service rather than health law.