Google Research introduced the Biomarker Discovery Framework, a multi-agent artificial intelligence system designed to automatically identify clinically meaningful biomarkers from continuous physiological signals captured by wearable devices. The framework addresses a critical bottleneck in health research: while wearable sensors now collect vast amounts of data from populations, converting this raw information into reliable biomarkers has remained a labor-intensive, error-prone process prone to spurious correlations and statistical oversights. The system combines deterministic statistical analysis with generative AI reasoning through a structured six-phase workflow involving specialized agents that handle hypothesis generation, data validation, statistical testing, adversarial checking, and literature synthesis. A human-in-the-loop architecture ensures researchers maintain oversight throughout discovery. The framework enforces strict statistical validity checks, including an 11-test adversarial filtering stage to catch potential leakage, overfitting, and confounding effects that traditional AI systems often miss. In testing across three large cohorts totaling 9,279 participant-observations, the framework identified 41 candidate biomarkers for mental health conditions and 25 for metabolic disease, recovering known clinical signals while discovering convergent findings across independent datasets. Google says the tool successfully balanced automated discovery with human expert judgment, positioning it as a model for how generative AI can accelerate scientific research while preserving methodological rigor.