Researchers propose AI governance model using compute budgets

A team of researchers from India's IIIT Bangalore and IIT Madras has proposed a novel governance model for AI agents called 'Resourced Authority', detailed in a new paper submitted to arXiv. The model introduces a mechanism-design framework where governance is exercised through controlled resource allocation—specifically, AI agents' compute budgets—rather than traditional command-and-control approaches.

The system works through a sequential game where human stakeholders contribute governance tokens to either support or reject an AI agent's operations. A funding aggregator converts these contributions into weighted support values, which pass through a two-threshold gate to determine authorization. Only authorized agents receive metered compute access via signed hardware licenses, effectively making compliance self-enforcing. The authors frame this as a 'compliance overlay' on AI deployments, positioning compute as the primary governance lever within a 'Safe AI' paradigm.