Develop a Standard Taxonomy for Compute Governance Capacities

Develop a standard, broadly accepted taxonomy for compute governance in artificial intelligence that formalizes core capacities—such as visibility, allocation, and enforcement—and maps associated mechanisms, to replace ad hoc classifications and enable consistent analysis and policy design.

Background

To organize how compute can be leveraged for AI governance, the paper introduces a bespoke taxonomy of three capacities: visibility (regulatory insight into development and deployment), allocation (steering compute toward beneficial outcomes), and enforcement (preventing or responding to rule violations).

The authors note that, despite related work, there is no standard taxonomy in the field. Establishing a common taxonomy would support coherence across research, policy discussion, and practical implementations of compute governance measures.