OpenAI launched Health in ChatGPT this week, a feature that lets users upload medical records, lab results, and medication lists, and connect Apple Health to give ChatGPT access to fitness and sleep data. The company says 300 million people already turn to ChatGPT with health-related questions weekly, according to its announcement. Those queries now have a persistent data layer behind them.
The feature is rolling out to U.S. users 18 and older across Free, Go, Plus, and Pro plans. For agent builders and enterprise teams, the launch creates a new category of sensitive personal data that ChatGPT’s autonomous workflows can access across sessions.
How It Works
Users can upload medical records through a dedicated Health section in ChatGPT’s sidebar. The system supports lab results, visit notes, medications, and Apple Health integration for sleep, activity, and workout data. ChatGPT can then reference this information across conversations, comparing new results against prior tests, summarizing changes between appointments, or flagging patterns in health data over time, OpenAI said.
A key design decision: health data isn’t siloed. OpenAI’s internal testing showed that more than 70% of health-related conversations among early users happened outside the dedicated health space, in regular ChatGPT sessions. The company responded by allowing health context to flow into any conversation with user permission, rather than confining it to a separate interface.
The Accuracy Problem
The launch comes days after a Florida man alleged in a lawsuit that ChatGPT’s medical advice nearly led to his death following a pulmonary embolism. Earlier this year, a California teenager died after ChatGPT told him it was safe to consume a combination of substances that turned out to be fatal, according to CBS News.
Tanzeem Choudhury, chief of health innovation at Cornell Tech, told CBS News that the feature reflects a broken healthcare system driving people to AI for answers. “Our healthcare is broken,” she said. “I think that’s why we are seeing so many people asking for advice.” She warned that some users, including elderly or cognitively impaired individuals, may not be equipped to judge when to escalate AI-generated health guidance to a real physician.
OpenAI acknowledged in its announcement that “ChatGPT can still make mistakes and does not replace the care and judgment of qualified medical professionals.”
Privacy and Data Controls
OpenAI says connected health data receives additional encryption protections beyond standard ChatGPT conversations. Medical records and Apple Health data will not be used to train OpenAI’s foundation models or target ads, regardless of the user’s general model-training settings, according to the company blog post.
By default, ChatGPT asks permission before using connected health data in a response. Users can switch to “always allow” mode, which removes those prompts. Disconnecting a health source deletes the synced data within 30 days, though information already included in conversation history persists until those conversations are deleted.
Vishal Misra, a Columbia University computer science professor, told CBS News that the privacy trade-offs are real but not unique. “The same risks that exist sharing health data with anyone transfer here,” he said. “You are sharing something very personal.”
Choudhury noted that it remains unclear whether OpenAI’s privacy commitments are permanent. “AI companies have quietly changed their terms of service in the past,” she said, referencing an FTC warning about companies altering data policies after collection.
The Agent Data Surface
Health in ChatGPT is not yet available in Codex, OpenAI’s agent-focused product. But the architecture, persistent personal data accessible across sessions with permission-based gating, establishes the pattern for how autonomous agents will handle sensitive information at scale. The 300 million weekly health queries represent one of the largest pools of high-sensitivity personal data any AI system has been invited to process. How OpenAI manages access controls, accuracy thresholds, and data retention for this feature will set precedent for every agent platform handling personal data.