AI Health Wearables
Not just measuring. Understanding.
A step counter measures. An AI health wearable does on-device inference — it classifies, patterns, and alerts based on a model running in the hardware. The distinction determines what the device can do for you offline, in real time, and without sending your biometrics to a server.
The definition
What makes a health wearable an AI wearable
The line is on-device inference. A health wearable runs a machine learning model locally — on the chip in the device — to classify or interpret sensor data.
An ECG that measures voltage is a sensor. An ECG that runs a classifier on the waveform and identifies atrial fibrillation locally — that is an AI health wearable. Apple Watch Series 4+ passed FDA clearance for exactly this in 2018. It was the first mass-market on-device medical AI at scale.
Categories
AI health wearable categories and key products
Smartwatches with health AI
Apple Watch Series 10, Samsung Galaxy Watch 7Features
- ECG (atrial fibrillation detection)
- Blood oxygen (SpO2)
- Crash detection (fall model)
- Temperature trend
- Irregular heart rhythm notifications
On-device AI
Apple Watch: S10 chip with on-device ECG classification. Crash detection runs a fall-detection model locally.
Most capable general-purpose health AI wearable. Requires pairing with iPhone for full functionality.
Dedicated health bands
Oura Ring Gen 4, Whoop 4.0Features
- Sleep staging (light/deep/REM classification)
- Readiness/recovery score
- Strain tracking
- Heart rate variability
- Body temperature monitoring
On-device AI
Oura runs sleep staging models on-device. Whoop runs strain/recovery models locally.
Purpose-built for continuous health monitoring. No display (Oura) or minimal display (Whoop).
Continuous Glucose Monitors (CGM)
Dexcom G7, Abbott Libre 3Features
- Real-time interstitial glucose readings
- Trend arrows (direction + rate of change)
- High/low glucose alerts
- Integration with insulin pumps
- Pattern analysis
On-device AI
Alert classification runs on sensor. Pattern analysis app-side. Dexcom G7 has a 30-minute warm-up vs 2-hour for previous generation.
FDA-cleared medical devices. Prescription required in the US for most CGMs.
Smart patches
Zio Patch (cardiac), various hydration patchesFeatures
- Extended ECG (up to 14 days)
- Arrhythmia detection
- Hydration monitoring
- Medication adherence tracking
On-device AI
Most cardiac patches store data locally and AI analysis occurs after retrieval. Frontier: real-time local analysis.
Generally prescription/clinical products. Consumer smart patches for wellness are early.
On-device vs cloud
Why on-device matters for health data specifically
| Aspect | On-device | Cloud |
|---|---|---|
| Alert latency | Immediate — model runs locally, alert fires in milliseconds | 1–30 seconds depending on network — unacceptable for critical health alerts |
| Privacy | Raw biometric data stays on device. Health data is categorically sensitive. | Biometric data transmitted and stored on servers. Subject to breach, policy change, and data sale. |
| Offline function | Works without network — important during exercise, travel, rural areas | Requires connectivity for AI features — degrades to raw sensor only offline |
| Personalization | Model can be fine-tuned to individual baseline — your normal HR, your sleep patterns | Population-level models — less personalized without significant cloud ML investment |
The frontier
Continuous context fed to a personal AI
Current AI health wearables measure and interpret one type of data — sleep, heart rate, glucose. The frontier is devices that do not just measure but understand: feeding continuous physiological context to a personal AI that knows your baseline, your patterns, and your goals.
Kin is benned's personal AI entity. Health wearables are a natural data source for Kin — continuous biometric context that a personal AI can reason about, surface relevant insights from, and act on without sending raw health data to a server.
Regulation
EU and US regulatory context for AI health monitoring
EU
The EU AI Act classifies AI systems in health monitoring by risk. High-risk AI (medical devices influencing treatment decisions) requires conformity assessment before deployment. Consumer wellness wearables below medical device thresholds face lower requirements.
GDPR applies to health data — explicit consent required, right to erasure, data minimization. On-device processing significantly simplifies GDPR compliance.
US
FDA regulates Software as a Medical Device (SaMD). Features like ECG atrial fibrillation detection and blood glucose monitoring require 510(k) clearance or De Novo authorization.
Consumer wellness features (sleep tracking, general fitness) typically fall under the FDA's general wellness policy and do not require 510(k) clearance.
FAQ
Common questions
What makes a health wearable an AI wearable?
On-device inference on biometric data. Not just measuring a value — running a classification or prediction model locally on what was measured.
Is Oura Ring an AI wearable?
Yes. Oura Ring runs sleep staging models, readiness scoring, and activity detection on-device without sending raw sensor data to the cloud for every decision.
Are there regulatory requirements for AI health wearables?
Yes. EU AI Act applies to health AI by risk tier. FDA SaMD framework applies in the US. Features like ECG atrial fibrillation detection and CGM require clearance. Consumer wellness features generally do not.
Last updated: July 2026