Apple Health AI Study app: write your AI data rules down
Apple launched a Health AI Study app with OpenEvidence: up to 200,000 US participants, 12+ months, queries deleted after 24 hours. Copy the data rules.
Apple launched a Health AI Study app with OpenEvidence on November 15. It is a research app, not a product. People ask health questions, an AI answers from clinical sources, and Apple studies how they use it. The interesting part for an operator is not the medicine. It is the list of data rules that Apple put in front of the feature: a deletion clock, a no-training promise and an age gate. Most small-business AI features ship with none of these written down.
What actually happened
TechCrunch reported the launch on Sunday:
- The app. Apple and OpenEvidence, a healthcare AI company, built a study app for health questions. The answers use clinical sources. TechCrunch names The New England Journal of Medicine, JAMA and Nature.
- The scale. Apple wants up to 200,000 participants. The study runs for more than 12 months.
- The gate. US users only. iPhone 15 or later, iOS 26 or later, Apple Intelligence on, and the user must be older than 21.
- The feedback loop. Participants rate answers, give feedback and get a survey each month.
- The optional data. With HealthKit access on, a participant can share resting heart rate, activity, sleep or lab results to get more specific answers.
- The data rules. Apple says it anonymizes data before it goes to a third-party cloud provider. Apple says the data will not train AI models. Apple says OpenEvidence deletes query and response data after 24 hours.
- What is missing. TechCrunch does not name the AI model or the cloud provider. It also says it is not clear if Apple will put AI answers into the Health app or Siri.
Apple's health lead, Sumbul Desai, said in a statement that responsible work in this area starts with "scientific rigor and clinical grounding."
Why the Apple Health AI Study matters for your business
You do not run a health study. But you probably have, or want, an AI feature that touches customer data: a support bot, a quote assistant, an intake form that a model reads. Apple shows the order of operations. It wrote the limits first and released the feature second, to a capped group, with a rating on each answer.
This matters more when the data is sensitive. We wrote about the gap when ChatGPT Health rolled out in the US: the law does not always follow the data into the AI tool. Your own written rules must do that work.
Four rules to copy this week:
- Set a deletion clock. Decide how long your AI vendor keeps prompts and answers. Apple's number is 24 hours. Find your vendor's number in the contract, not in the marketing page.
- Get the no-training term in writing. If the vendor can train on your customers' text, say so to customers or change the plan.
- Ground the answers. Apple limits the model to named sources. Limit yours to your price list, your policies and your product data.
- Start capped, and measure. Release to a small group. Put a thumbs up and thumbs down on each answer. Read the bad ones each week.
Key takeaways
- Apple launched a Health AI Study app with OpenEvidence on November 15, for US users
- The study targets up to 200,000 participants and runs for more than 12 months
- Apple says OpenEvidence deletes query and response data after 24 hours, and that the data will not train AI models
- Participants must be older than 21 and use an iPhone 15 or later with iOS 26 or later
- The AI model and the cloud provider are not named; treat the privacy terms as Apple's statements
Your AI feature needs a retention rule before it needs a launch date. We build customer-facing AI with the deletion window, the no-training term and the answer sources set in code and in the contract. See how we build AI features you can explain to a customer, or tell us what data your assistant will touch.
Sources: TechCrunch.
- #apple
- #openevidence
- #health-ai
- #data-retention
- #ai-privacy
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Qwen4-Max released: $1.50 input, cost per task down 62%
Alibaba released Qwen4-Max with a 1M-token context and API access at $1.50 input and $4.50 output. It is not open weights. Measure cost per task before you switch.
Read itAI accounting software: Blue J $176M, Sumary $25M
AI accounting software raised two large rounds: Blue J $176M for tax research and Sumary $25M for firm workflows. Ask your accountant which AI reads your books.
Read it