AI Model Registries: A Foundational Tool for AI Governance

National AI model registries could provide governments with visibility into frontier AI development comparable to their oversight of other high-risk industries like aviation and pharmaceuticals, helping policymakers understand what powerful AI systems exist and who is developing them.

Key information to collect in registries should include model architecture, size, compute resources used, training data, and safety evaluation results, enabling regulators to identify potentially high-risk systems while the framework is designed to protect developers' intellectual property and sensitive information.

Compliance could be encouraged through "injunctive action"—requiring third parties to only use registered models—rather than relying solely on direct penalties, which the authors suggest would be more efficient and create stronger incentives for voluntary participation in the registry system.

Registry systems can balance innovation with safety by providing governmental insight into AI development without creating excessive regulatory burden, drawing lessons from how other industries manage similar oversight challenges while maintaining competitive markets.

Timely and accurate registration of AI models is presented as foundational to responsible AI governance, allowing governments to monitor emerging risks and support safety measures before advanced systems are widely deployed.