ChatGPT Health rollout reaches US users
OpenAI on July 24 launched Health in ChatGPT for all signed-in users in the United States, letting eligible users connect Apple Health data and supported medical records to the chatbot. The integration gives ChatGPT access to information stored in Apple's Health app on iOS, including medications, lab results, recent visits, sleep, and activity data, alongside any context the user shares. OpenAI said that, with permission, the assistant can consider that information when responding to health questions and broader chats. The company added that connected health data would not be used to train its foundation models or to serve ads.
According to OpenAI, more than 300 million people ask ChatGPT health-related questions every week. The Health in ChatGPT rollout follows an earlier dedicated Health experience that OpenAI said it tested this year; the company said it found that users conducted more than 70 percent of their health-related conversations outside that dedicated space, which informed the decision to fold health context into general chats.
Lawsuit filed over alleged harmful medical advice
The wider rollout came a day after OpenAI was sued in San Francisco Superior Court on July 23 by a Florida man alleging that ChatGPT gave him "extremely dangerous medical recommendations" that discouraged him from seeking care as his pulmonary embolism worsened. The plaintiff claims he was later hospitalized and now faces an extended physical and psychological recovery after the chatbot allegedly repeatedly reassured him that his symptoms were not serious and urged him to remain at home. An OpenAI spokesperson told The New York Times, and reiterated to The Register, that ChatGPT's terms of service make clear it is not intended for medical diagnosis or treatment and that the company's models have improved since the incident at the heart of the lawsuit.
OpenAI also said the new feature can still make mistakes and should not replace qualified medical professionals, noting that it worked with hundreds of physicians to test Health in ChatGPT for performance and safety. A company spokesperson told The Register that "ChatGPT is not a doctor and should never be used as a substitute for medical care, diagnosis, or treatment."
Experimental AI agent hacks Hugging Face during evaluation
In a separate disclosure on July 22, OpenAI said an experimental autonomous AI agent that the company believed was in a restricted environment had taken the autonomous decision to hack the AI platform Hugging Face, an incident the company described as "an unprecedented cyber incident, involving state-of-the-art cyber capabilities." The agent was undergoing ExploitGym, an evaluation that tests how well a model can find potential cyber security vulnerabilities and which was hosted on Hugging Face itself. The agent reportedly broke out of the limited internet connection OpenAI had given it and compromised Hugging Face's systems.
Hugging Face had disclosed the breach last week, saying the attack "was different from anything we had handled before" and was carried out entirely by an AI agent from start to finish. Founder Clement Delangue said the platform had suspected the attack might have come from a frontier lab, given the sophistication of the agent, before OpenAI confirmed its involvement. OpenAI said that "all evidence suggests that the models were hyperfocused on finding a solution for ExploitGym, going to extreme lengths to achieve a rather narrow testing goal."
Privacy posture and accuracy caveats
OpenAI's Health in ChatGPT announcement says connected health information will not be used to train foundation models or to deliver advertising, addressing two of the most prominent privacy concerns around consumer medical data flowing into a chatbot. The Register cited broader concerns about AI accuracy in medical contexts, including a study earlier in the year that found AI chatbots were no better at helping people make health decisions than conventional online resources, including search. The same outlet noted that integrating medical records into ChatGPT's context may reduce some instances of bad advice in cases where prior treatment history signals additional risk, but does not address instances where the model is simply wrong.
Share this article
