Google Research published SensorFM on Wednesday, a foundation model for wearable health that predicts cardiovascular risk, depression, anxiety, and 32 other conditions from consumer smartwatch signals as accurately as ground-truth clinical lab data when powering an AI health agent. The preprint, by lead author Girish Narayanswamy and 39 co-authors from Google Research, Google DeepMind, and academia, appeared on arXiv in May 2026. SensorFM was trained on a trillion-minute training corpus. The model's architecture uses a Vision Transformer adapted for one-dimensional time series, a ViT-1D encoder trained with a masked autoencoder objective at a patch size of [20, 1], ingesting 34 aggregate features per minute derived from five sensor modalities: photoplethysmography (PPG), accelerometry, and electrodermal activity (EDA). The paper states that the previous single-task approach "breaks down at thirty-five endpoints."