Google's SensorFM Wants to Be One AI Brain for All Your Wearable's Health Data

Google Research has introduced SensorFM, an AI model that learns to understand the messy stream of data coming off your smartwatch — heart rate, sleep, movement, skin temperature — all at once, instead of one feature at a time. It was trained on more than a trillion minutes of anonymized sensor data from five million Fitbit and Pixel Watch users across over 100 countries, which Google says makes it the largest and most diverse wearable dataset ever used for this kind of model.

Here’s the problem SensorFM is trying to solve. Today, every health feature on your watch is its own little program: one model detects sleep stages, another estimates heart health, a third guesses your stress level. Each needs its own expensive, hand-labeled training data. SensorFM is a foundation model — the same idea behind ChatGPT, but for body signals instead of text. It learns general patterns of human physiology from raw, unlabeled data, and can then be adapted to many different health questions with relatively few examples.

The training trick is simple to picture: the model gets sensor recordings with chunks deliberately hidden, and learns to fill in the blanks. Since real wearable data is full of genuine gaps — you take the watch off, the sensor slips — the researchers taught it to handle both kinds of missing data. In tests on data from 13,985 people it had never seen, SensorFM beat specialized comparison models on 34 out of 35 tasks, covering sleep, heart and metabolic health, and even mental-health indicators like depression and anxiety symptoms.

What’s behind this? Google is laying the groundwork for AI health assistants that actually know something about you. A chatbot that can read your last month of sensor data as fluently as it reads text could give far more personal answers than one that just knows general medical facts. A fair caveat: this is research, not a product, and predicting anxiety symptoms from a wrist sensor is exactly the kind of claim that needs real-world validation — and careful privacy rules — before it lands on anyone’s wrist.

What this means for you: Nothing changes on your watch today. But if you own a Fitbit or Pixel Watch, this is the direction your data is feeding into: fewer isolated features, more of a single health layer that AI assistants can draw on. Worth watching — and worth keeping an eye on the privacy settings when it arrives.

Sources

• Google Research: SensorFM blog post
• The Decoder: Google’s SensorFM turns messy wearable sensor data into a general-purpose health intelligence layer