Paw posture is a robust indicator for injury, pain, and age.

Inferring biological states from animal behavior is a crucial but challenging step in biomedical discovery that is constrained by variability and labour-intensive assays, even with AI-powered tools. Here, we show that simple images of static paws, analyzed by our custom keypoint segmentation AI-tool provide accurate read-outs for a wide array of physiological states. Without invasive testing, our method detects postural changes associated with nerve injury, acute pain, aging, and the genetic loss of kpna4, a regulator of paw innervation. Leveraging the toe-spread-reflex, a spinal-circuit driven response, the approach requires no habituation and shows low behavioral variability. Individual digits emerge as biomarkers for internal states with digit V indicating neuropathic pain during nerve damage and digit I reflecting loss of kpna4. Our model is freely available and can readily be adapted to other tasks or species. These findings establish unstimulated paw posture as a scaleable, low-cost, readout for biological states.