Develop regulatory frameworks for modern AI systems

Determine whether and how governments should regulate the development and deployment of frontier foundation models built in private laboratories, specifying the scope and mechanisms of regulatory access needed to evaluate safety and performance when independent replication is infeasible.

Background

Unlike traditional industries where products can be tested independently, frontier AI models cannot be readily replicated or evaluated without access to proprietary training data and procedures. The authors argue that regulators and academics need access to information considered proprietary, yet acknowledge that policymakers lack clarity on appropriate regulatory approaches in this domain.

References

In the case of modern AI, however, governments do not know if and how they should regulate.

— An Economy of AI Agents  (2509.01063 - Hadfield et al., 1 Sep 2025) in Institutions for AI agents, Subsection “Rethinking the legal boundaries of the corporation”

This is the open gap: consent to training has been established in several places, but governance of the trained model has not been well established.

— Govern the Model, Not Only the Data: Storage, Circulation, and Learning in Creative AI  (2609.03800 - Perry et al., 3 Sep 2026) in Section 3.2, “Learning: consent established, model governance open”