Last week, the idea of an AI that knows your health stopped being a demo. Two of the largest companies in consumer technology opened health assistants to the public within three days of each other. Neither is available in Switzerland or the rest of Europe yet. Both tell us a lot about where this is going, and about the questions everyone should ask before handing their health data to a chat window.
What happened
On 21 July, Samsung announced Samsung Health Assistant, an AI assistant inside the Samsung Health app, in beta for eligible users in the US. It connects sleep, activity, nutrition, mindfulness and vitals, explains how they affect each other, and answers questions with guidance Samsung says was validated by physicians and certified health coaches. Samsung is clear that it "will not give medical advice".
Two days later, on 23 July, OpenAI made ChatGPT Health available to every adult in the US, on every plan, on the web and on iPhone. People can connect Apple Health, MyFitnessPal and Function, and link medical records through systems such as Epic and Oracle Health. According to the company, health questions in ChatGPT grew from 230 million a week when testing began to 300 million now. OpenAI says it does not use this data to train its models, and its terms say the service is not meant to diagnose or treat anything.
When ChatGPT Health first launched in January, people in the European Economic Area, Switzerland and the UK were left out of the rollout, and the medical record connections are US only. That is still the case today.
Why it matters
Three hundred million health questions a week is not a fad. It is a signal that people want one place that can make sense of their own numbers, in plain language, at the moment they have the question. We understand that wish very well. It is the gap Kata was built for: one conversation for the whole day, with your sleep, food, training and recovery behind it.
But a health assistant is only as good as the trust you can put in it, and trust depends less on how clever the answers are than on what happens to everything you tell it. Your sleep, your weight, your cycle, your medication, what you ate at midnight: this is the most personal data most of us produce. Once it is connected, it is very hard to take back.
Five questions to ask before you connect anything
Whichever assistant you try, now or when these services reach Europe, these are the questions that tell you more than any feature list.
- Where does my data live? On your phone, or on the company's servers? If it is on their servers, in which country, and for how long?
- What happens to what I type? Every AI reply has to be written somewhere. Ask what travels, whether it is encrypted, and whether a copy is kept afterwards.
- Is it used to train anything? And if the answer is no, is that a promise in the privacy policy or a setting you have to find and switch off?
- Can I really delete it? Look for a clear way to remove your data and your account, and for what deletion actually covers.
- Does it know its limits? A good assistant says plainly when a question belongs with a doctor, and does not dress a guess up as a diagnosis.
None of these questions is a trick. A service that takes your health seriously will have short, clear answers to all five.
How we answer them at Kata
We built Kata around those questions before we wrote the first line of code. Your health data, your chats and Kata's memory of you live on your phone, not on our servers. When you ask Kata something, only your message and the context it needs travel, encrypted, to the AI to write the reply, and Kata keeps no copy. There are no ads and nothing is sold. And Kata AI is a companion for everyday health, not a doctor: it coaches, it does not diagnose, and it tells you when a professional is the right next step.
We think this is the only way a health assistant can work over years rather than weeks. A coach is useful when you can tell it everything, and people only tell everything to something that keeps it with them.
The short version
AI health assistants are arriving quickly, and some of them will be genuinely useful. Before you connect your records to any of them, ask where your data goes. The answer tells you what kind of relationship you are starting.
