Detecting AI Training via Datacentre Power Cadence Tracking
October 23, 2026
A Viterbi tracking method detects synchronous model training by monitoring facility-level power fluctuations. The tracker identifies at least 96% of training runs against inference workloads with a 1% false-alarm rate when sampling at 2 Hz or faster.
HOW THIS AFFECTS YOU
●
policyYou can use power monitoring to verify compliance with AI governance agreements without trusting cluster operators.