Samsung Tyrant Clause: They provide your personal health data for AI training—if you don’t agree, they will wipe all your sleep and menstrual period records clean.

Samsung requires Samsung Health users to consent to using sleep, medication, medical history, and menstrual cycle data for AI training; if users do not agree, data synchronization will be terminated and already backed-up health records will be deleted, like a霸王條例.
(Background recap: Chrome browser store will roll out new policies in August: banning plugins related to prediction markets, AI extensions, data collection, and more)
(Additional background: Grok Build was hit by reports of secretly uploading users’ “entire home directory” to the cloud, startling developers: everything has leaked)

Table of contents

Toggle

• “Consent” is redefined
• Four types of data—who’s watching
• Vitals needs data to grow

Korean tech giant Samsung has, starting this week, begun showing a consent form inside its Samsung Health app: once you turn on the app, you’ll be asked to agree to “use sleep, medication, medical history, and menstrual cycle data for AI training and modelling.”

If you don’t agree, the backup feature will be shut off immediately, and health records already stored on Samsung’s servers will also be deleted in bulk. Samsung packages this setting as a user “choice,” but when the only two options are “hand over your data” or “lose your data,” do users still really have a choice?

“Consent” is redefined

Deep in Samsung Health settings, a new toggle has been added: “Consent to the Use of Health Data for AI training and modelling.”

Turning it on means authorizing Samsung to use personal health metrics for model training and algorithm improvements; turning it off will trigger an immediate in-app warning, telling users they will be unable to sync health data to their Samsung account, and that existing data will be deleted, unless required to be retained by law—once the legal retention period expires, it will be cleared anyway.

In short, this is no longer an option about whether to use a particular new feature, but an option about whether to keep your own data.

Four types of data—who’s watching

Samsung plans to collect four categories of data: sleep, medication, medical history, and menstrual cycle tracking. These are all private information either actively entered by users or passively detected by devices. The level of detail far exceeds ordinary step counts or heart rate; menstrual cycle tracking and medical history data in particular involve highly sensitive bodily privacy. In its official description page for Samsung Health, Samsung states that these data will be used to “improve Samsung Health,” using machine learning algorithms to analyze health status and optimize model accuracy.

The more sensitive part is that Samsung also admits that some of the collected data may be manually reviewed—by Samsung employees, or by third-party contractors. That means the data users provide isn’t just raw input fed to algorithms; it also includes the possibility that real people will look at it. Compared with Samsung’s privacy-protection posture emphasized on its description page, this line reads especially jarring: it says it wants to protect privacy while also admitting that someone will be watching.

When most users check “agree,” they’re likely thinking about new features enabled by wearable devices, and they may not particularly realize their medical history and menstrual cycle records could appear on a reviewer’s screen. As a result, many outside observers generally believe this isn’t so much a privacy upgrade as treating users’ bodily data as free training material for models.

Vitals needs data to grow

The Samsung Health app has just completed a large-scale generative AI overhaul, launching alongside the Galaxy Watch 9 and One UI 9 Watch. In its official announcement, Samsung introduces a new tool called Vitals: it compares the body signals users generate overnight against their personal baseline values, monitoring five indicators—heart rate, heart rate variability, respiration rate, skin temperature, and blood oxygen.

Heart rate variability, simply put, is the subtle fluctuation in the intervals between heartbeats; the higher the value, the better your body’s recovery capability typically is. Once any of these signals show abnormalities, the app will proactively pop up alerts, giving an early warning of possible illness or fatigue.

But the shared prerequisite for these features is that the model needs a large amount of real health data to be calibrated accurately enough. Users’ real-time physiological records are the fuel for this set of algorithms. Trading the quality of future features for today’s data authorization: the pricing power for this deal is never in the hands of users.