OpenAI’s Pivot to Clinical Data Lacks Evidence, Arrives Amidst Safety Lawsuit

OpenAI is attempting to move ChatGPT from a general-purpose chatbot to a personalized health tool, but the timing and lack of transparent safety data are alarming. On July 23, 2026, the company launched "Health in ChatGPT," a feature allowing U.S. users aged 18 and older to connect the chatbot to Apple Health and supported medical records. The goal, according to OpenAI, is to provide "more informed, personalized conversations" by giving the AI access to lab results, medications, and activity data.

However, the timing of this rollout is a massive red flag. As reported by The Register, OpenAI was sued on Wednesday in San Francisco Superior Court by a Florida man who alleges the chatbot provided "extremely dangerous medical recommendations." The plaintiff claims the AI repeatedly reassured him that his symptoms were not serious and urged him to stay home, which discouraged him from seeking care as a pulmonary embolism worsened. The lawsuit alleges the man now faces years of intensive physical and psychological recovery after being hospitalized.

Despite this, OpenAI is pushing further into the clinical space. The company claims it worked with hundreds of physicians to test the feature for performance and safety and asserts that its models—specifically GPT-5.5 Instant and GPT-5.6 Sol—have made "meaningful gains" in recognizing when urgent care is necessary. Yet, these claims are not backed by peer-reviewed data. Instead, the company relies on internal testing and anecdotal testimonials from early users, such as a technical program manager named Blake.

***

**Opinion:** *OpenAI is treating the human body as a beta test. By integrating real-time medical records without publishing rigorous, independent safety studies, the company is gambling with patient lives. The fact that this feature launched a day after a lawsuit alleging near-fatal medical advice suggests a corporate culture that prioritizes feature velocity over patient safety.*

***

The broader track record for AI in medicine is equally concerning. The Register cites a study from earlier this year finding that AI chatbots were no more effective at helping people make health decisions than standard search engines. Another study found that Large Language Models (LLMs) failed at early differential diagnosis in over 80% of cases. Furthermore, the Auditor General of Ontario found that AI note-taking systems used by doctors frequently hallucinated information and introduced medication errors into patient notes.

OpenAI maintains that ChatGPT is not a doctor and is not intended for medical diagnosis or treatment. An OpenAI spokesperson told The New York Times and The Register that the chatbot's terms of service make this clear. While OpenAI claims it will not use connected health data for advertising or to train foundation models, the fundamental risk remains: the bot can still be wrong. As The Register notes, while access to medical records might prevent some errors, it does not stop the AI from hallucinating or providing dangerous advice.