Apple and OpenEvidence Will Test Medical AI on 200,000 iPhone Owners for a Year
Apple and OpenEvidence have launched an app in the US where an AI answers health questions and people rate the answers. Up to 200,000 owners of iPhone 15 and newer can take part, and the study will run for more than 12 months. For example, a person asks about fatigue, gets an answer with links to Nature, rates it, and once a month completes a survey. For product teams, the takeaway is this: trust in medical AI is being tested on real people, not just in benchmarks.
AI-processed from RB.RU; edited by Hamidun News
People have long been asking AI systems what to do about blood pressure, test results, and back pain. Together with the medical AI platform OpenEvidence, Apple has launched a research app in the US to find out how people perceive such advice. Up to 200,000 owners of iPhone 15 and newer can take part, and the study will run for more than 12 months.
Picture this: a person asks why they are constantly tired, gets an answer in plain language with a link to a scientific journal, rates it, and a month later answers a survey on whether the advice helped. This is exactly what Apple wants to measure: whether the answers are understandable and how much people trust them.
What Apple Launched
According to TechCrunch, the app has been available to US users since November 15, 2026. To participate, you must be over 21, have an iPhone 15 or newer, run iOS 26 or later, and turn on Apple Intelligence, Apple's own set of AI features.
The mechanics are simple. A user asks a health question, and the AI answers based on scientific sources: The New England Journal of Medicine, the Journal of the American Medical Association, and Nature. The person then rates the answer and leaves feedback. Once a month, participants will be invited to take a survey so Apple can see how attitudes toward AI-generated medical information change.
Another feature is preparing for a doctor's visit. If a question requires medical care, the service directs the person to a relevant specialist. This is an important limitation: the app is positioned as an assistant and reference tool, not as a digital doctor.
Personalization via HealthKit and Privacy
Participants can optionally share data from HealthKit, Apple's platform for storing health information. This includes resting heart rate, physical activity, sleep, and lab results. With such data, the advice can become more personal: an answer about fatigue can take into account how the person has been sleeping over recent weeks, not just general recommendations from an article.
Health is one of the most sensitive categories of data, so Apple set clear boundaries right away. Before data is passed to a secure third-party cloud provider, it is anonymized. It will not be used to train AI models. OpenEvidence deletes queries and answers after 24 hours. Apple's Vice President of Health and Fitness, Sumbul Desai, said the study will help understand how people interact with medical AI services, understand their answers, and evaluate the information they receive in everyday life.
What This Means for Developers and Product Teams
First, Apple is measuring not only model quality but also user trust. For medical products, this often decides everything: even an accurate answer is useless if the person does not understand it or does not believe it.
Second, answers with links to peer-reviewed journals are a pattern that can be adopted in other high-stakes fields such as finance, law, and education. A verifiable source reduces the risk of fabricated facts and gives the user something to stand on.
Third, privacy is built into the design: anonymization, a ban on training models with the data, and fast deletion of queries. If you work with medical or other sensitive data, such a scheme is worth planning from day one.
There are limitations too. The study runs only in the US and only on newer iPhones, so the sample will be narrow in terms of countries and devices. No results have been published yet, and Apple has not said when it will publish them. The company also has not specified whether it will add medical answers to the Health app or to Siri.
The signal is clear, though. Apple, OpenAI, and Anthropic are all seriously working on medical questions: according to TechCrunch, earlier in 2026 OpenAI reported that users ask its services more than 300 million health questions every week. Whoever first figures out which phrasings people find understandable and worth trusting will gain an advantage in one of the most sensitive segments of the AI market.
- 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 Launched?
According to TechCrunch, the app has been available to US users since November 15, 2026. To participate, you must be over 21, have an iPhone 15 or newer, run iOS 26 or later, and turn on Apple Intelligence, Apple's own set of AI features. The mechanics are simple. A user asks a health question, and the AI answers based on scientific sources: The New England Journal of Medicine, the Journal of the American Medical Association, and Nature. The person then rates the answer and leaves feedback.
What are the key details on Personalization via HealthKit and Privacy?
Participants can optionally share data from HealthKit, Apple's platform for storing health information. This includes resting heart rate, physical activity, sleep, and lab results. With such data, the advice can become more personal: an answer about fatigue can take into account how the person has been sleeping over recent weeks, not just general recommendations from an article. Health is one of the most sensitive categories of data, so Apple set clear boundaries right away.
What This Means for Developers and Product Teams?
First, Apple is measuring not only model quality but also user trust. For medical products, this often decides everything: even an accurate answer is useless if the person does not understand it or does not believe it. Second, answers with links to peer-reviewed journals are a pattern that can be adopted in other high-stakes fields such as finance, law, and education. A verifiable source reduces the risk of fabricated facts and gives the user something to stand on.
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.