Hackers Could Turn AI Training Jobs Into Weapons Against the Power Grid

The report describes Bit2Watt, a cyber-physical research finding showing how synchronized AI/GPU training jobs can be abused to create high-frequency power demand swings that induce voltage fluctuations, harmonics, and reduced damping in local grids and data-center equipment. Researchers demonstrated the effect in simulations and hardware experiments, noted the threat can operate through legitimate user-level controls (making it stealthy to cloud monitoring), and recommended combining cloud workload telemetry with electrical monitoring, limiting workload synchronization, and coordinating scheduling with power engineers.