Google Research presents SensorFM

Google Research presents SensorFM for wearable health data

Researchers say the foundation model was pre-trained on more than one trillion minutes of de-identified wearable data from five million consented participants.

By AI Wire Editorial TeamPublished July 9, 2026Checked July 18, 20263 min read

Source brief: This page summarizes and attributes the primary material linked below. It is not independent confirmation of the organizations’ claims.

Key facts

Status: Quick Brief
Coverage: Research
Primary record: 1 source
Last checked: July 18, 2026

What the source claims

The following points are attributed to the organizations in the source record; AI Wire has not independently reproduced them.

Google researchers report that SensorFM was pre-trained on more than one trillion minutes of de-identified wearable data from five million consented participants.
The researchers say the model transfers across 35 wearable-data prediction tasks.

What remains unknown

Clinical usefulness, subgroup performance, privacy protections, and performance outside the reported research setting.
Whether independent teams can reproduce the reported transfer results.

Topics in this brief

Research
Wearables
Health AI

What to watch

Peer-reviewed details, subgroup analyses, independent replication, and prospective clinical evaluation.

Sources and evidence

primary source

Google Research — SensorFM ↗

Accessed 2026-07-18. Claims above remain attributed to this source rather than presented as independent verification.

Corrections and updates

No corrections have been issued for this brief.

Last checked: July 18, 2026.

Request a correction →