Compute governance is the set of policies, controls, and oversight mechanisms applied to the large-scale computing hardware used to train and deploy advanced AI systems. Because frontier AI capability is tightly coupled to access to specialised accelerators, governing compute offers a measurable, supply-chain-anchored lever for AI policy. Mechanisms include export controls, usage reporting thresholds, and on-chip verification of training runs.
Content
- Because accelerators are physical, excludable, and produced by a concentrated supply chain, they are more governable than algorithms or data. Proposals span export controls, compute reporting thresholds, and hardware-enforced attestation of who ran what, balancing safety against innovation and concentration concerns.