Apple and OpenEvidence to Test Medical AI Chat With 200,000 iPhone Users
Apple has released a US app where AI answers health questions much like ChatGPT, but grounds its replies in clinical journals such as the New England Journal of Medicine. It was built with OpenEvidence, a company that develops AI for medicine. Apple wants to enroll up to 200,000 iPhone owners and study for more than a year whether the answers are useful. The takeaway for anyone building medical AI: test with real people first, then integrate into the product.
AI-processed from Habr AI; edited by Hamidun News
People already ask chatbots about their pulse, lab results and medications, but no one really knows whether such answers can be trusted. Apple decided not to release yet another medical bot right away. Instead, it wants to first test how this works in real life. To do that, the company has launched a separate research app together with OpenEvidence, a firm that builds AI for medicine and answers questions based on clinical sources. The plan is to enroll up to 200,000 iPhone owners in the US and observe them for more than 12 months. In short, Apple is collecting data on how useful medical AI is and whether it earns trust before embedding it in its own services.
What Apple has launched
Outwardly, the app looks like ChatGPT or Claude: the user types a health question and gets an answer. The difference lies in the sources. The answers draw on clinical data from The New England Journal of Medicine, the Journal of the American Medical Association and Nature. Apple says the answers will be simplified so that someone without a medical background can understand them.
For example, a person notices that their resting pulse has risen above normal and asks whether they should be worried. If they have granted access to HealthKit (the health data store on iPhone), the app can take into account their resting heart rate, activity level, sleep and even lab test results. The answer will then be more personal. The app also helps people prepare for a doctor's visit. For serious problems, it directs users to appropriate specialists.
How the study is set up
Only US residents aged 21 and older can take part. Participants need an iPhone 15 or newer, iOS 18 or newer, and Apple Intelligence turned on. The cap is up to 200,000 people, and the duration is more than 12 months. Participants rate the answers, leave feedback and complete a survey once a month. Through this, Apple wants to understand not only whether the answers are correct, but also how people perceive them: what seems helpful and what breeds distrust.
Apple's Vice President of Health and Fitness, Sumbul Desai, describes health questions as a deeply personal sphere and considers a dedicated study the most thoughtful way to examine how people interact with medical AI. According to her, the results will help broaden the scientific discussion.
Privacy and limitations
The most sensitive issue here is data. According to Apple, information is anonymized before being passed to a third-party cloud provider, and it will not be used to train any AI models. OpenEvidence deletes search queries and responses after 24 hours. That sounds strict, but it cannot yet be independently verified: everything rests on the companies' promises.
There are other limitations. The app is available only in the US and only on newer devices, so the sample will skew toward owners of current iPhones. Apple has not said whether such answers will appear in the Health service or in Siri. For now, this is a study, not a finished feature. Simplified answers do not replace a doctor, and the app itself suggests consulting a specialist.
In 2025, Apple already ran a Health Study examining how iPhone, AirPods and Apple Watch affect physical and mental health. The new project continues that line, but now with AI at its center.
What this means for developers and product managers
For those building AI products in sensitive areas, there are three practical lessons. First, tie answers to verifiable sources rather than to the model's general knowledge. Second, ratings of answers and regular surveys are needed as data for tuning the product, not as a formality. Third, build privacy into the architecture, with anonymization and short retention periods for queries.
If Apple later builds such answers into Health or Siri, it will already have data on the usefulness and trustworthiness from hundreds of thousands of people. That is a notable advantage over competitors who release medical bots without such testing.
- Nov 21, 2026OpenAI GPT-6 Cyber Found About 35 Flaws in iPhone, Mac and Other Apple Devices
- Nov 21, 2026Apple's macOS 27.3 beta 2 will let you limit AI agents' access to Mac files
- Nov 21, 2026Apple Is Building an AI Agent for Mac While Making It Harder for Outsiders to Get In: 14 Clicks Instead of 4
- Nov 21, 2026Apple Is Recruiting 200,000 Volunteers to Test a Health AI Chatbot
Key Questions Answered
What Apple has launched?
Outwardly, the app looks like ChatGPT or Claude: the user types a health question and gets an answer. The difference lies in the sources. The answers draw on clinical data from The New England Journal of Medicine, the Journal of the American Medical Association and Nature. Apple says the answers will be simplified so that someone without a medical background can understand them. For example, a person notices that their resting pulse has risen above normal and asks whether they should be worried.
How the study is set up?
Only US residents aged 21 and older can take part. Participants need an iPhone 15 or newer, iOS 18 or newer, and Apple Intelligence turned on. The cap is up to 200,000 people, and the duration is more than 12 months. Participants rate the answers, leave feedback and complete a survey once a month. Through this, Apple wants to understand not only whether the answers are correct, but also how people perceive them: what seems helpful and what breeds distrust.
What are the key details on Privacy and limitations?
The most sensitive issue here is data. According to Apple, information is anonymized before being passed to a third-party cloud provider, and it will not be used to train any AI models. OpenEvidence deletes search queries and responses after 24 hours. That sounds strict, but it cannot yet be independently verified: everything rests on the companies' promises. There are other limitations. The app is available only in the US and only on newer devices, so the sample will skew toward owners of current iPhones.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.