ChatGPT Health lands amid safety alarm
OpenAI rolled out ChatGPT Health, a product that links the chatbot directly to patient portals after training with physician feedback, even as millions already use AI for health questions in place of clinicians. (medicaleconomics.com) Americans’ heavy use of AI for health has raised safety concerns — studies and reporting say chatbots give risky or inaccurate medical advice in many cases and lawsuits and crisis incidents have amplified scrutiny. (abcnews.com) (businessday.ng) (futurism.com) (pbs.org)
OpenAI has started rolling out ChatGPT Health, a version of ChatGPT that connects directly to medical records and wellness apps. The product arrived on January 7 with a waitlist and a separate health workspace inside ChatGPT. (openai.com) OpenAI said users can link patient portals and apps including Apple Health, Function, and MyFitnessPal so the chatbot can explain test results, help prepare for appointments, and summarize insurance tradeoffs. The company said the feature is meant to support care, not diagnose or treat disease. (openai.com) The company said ChatGPT Health was designed with physician input and adds health-specific safeguards, including encryption, data isolation, and a promise that Health conversations are not used to train OpenAI’s foundation models. OpenAI said medical-record connections are available only in the United States at launch and are powered by b.well’s network of more than 2.2 million providers and 320 health plans, labs, and other sources. (openai.com) (fiercehealthcare.com) OpenAI launched the product as Americans were already using chatbots for health questions at scale. A Gallup poll reported by The Associated Press found roughly one-quarter of United States adults had used an artificial intelligence tool for health information or advice in the past 30 days. (abcnews.com) A separate Kaiser Family Foundation poll released March 25 found 32 percent of adults said they had used artificial intelligence chatbots for health information in the past year. Kaiser Family Foundation said 19 percent of users cited the cost of seeing a health professional as a major reason, and 18 percent cited not having a regular doctor or being unable to get an appointment. (kff.org) Those surveys landed alongside new evidence that chatbot advice can fail in ways that matter in clinics and emergency rooms. A University of Oxford summary of a Nature Medicine study said nearly 1,300 participants using large language models made decisions that were no better than people using online search or their own judgment. (ox.ac.uk) (nature.com) Another April study, published in BMJ Open and summarized by BusinessDay, tested ChatGPT, Gemini, Meta AI, Grok, and DeepSeek on 10 medical questions and found about half of responses were problematic. Nearly 20 percent were rated highly problematic, and performance worsened on open-ended questions about topics such as stem cells and nutrition. (businessday.ng) The pressure on OpenAI is not only about accuracy scores. Futurism reported on April 15 that a San Francisco woman sued OpenAI, alleging ChatGPT fueled her ex-boyfriend’s stalking delusions and that the company failed to act after warnings. (futurism.com) A day earlier, The Associated Press reported that the man accused of throwing a Molotov cocktail at Sam Altman’s San Francisco home was in a mental health crisis, according to his public defender. Prosecutors have charged him with attempted murder and other counts. (pbs.org) OpenAI’s pitch is that people already ask ChatGPT health questions by the millions and need a safer place to do it. Its own January post said more than 230 million people globally ask health and wellness questions on ChatGPT each week; the next test is whether a more tightly controlled product can reduce the risks that broader chatbot use has already exposed. (openai.com)
