Policy & Governance

Building the Legal Infrastructure for Safe and Beneficial AI

Developing practical, evidence-based approaches to AI governance that can adapt to rapidly evolving technology while ensuring safety and democratic accountability.

AI requires dynamic governance

I develop practical proposals for governing AI, especially to ensure AI is safe and beneficial for all. My work focuses on building the legal and regulatory infrastructure that will enable us to develop and adapt necessary rules for AI in a dynamic and evidence-based way.

In my 2017 book Rules for a Flat World: Why Humans Invented Law and How to Reinvent It for a Complex Global Economy, I addressed the critical challenge of developing appropriate law and regulation for a fast-moving, complex and global technology. I developed a proposal for how we recruit private sector efforts and investment to the challenge of devising more effective and responsive methods for aligning AI systems with democratically-set goals and principles. This approach, which I call regulatory markets, envisions the development of a robust sector of licensed private regulatory companies (profit and non-profit) that achieve publicly-set regulatory targets through innovation of new regulatory technologies and methods.

Regulatory markets envision a robust sector of licensed private regulatory companies that achieve publicly-set regulatory targets through innovation of new regulatory technologies and methods.

Adapting our legal infrastructure for AI systems

With Tino Cuéllar and Tim O'Reilly, I focus on the transformation of legal infrastructure to fit the complexities of AI systems. Governments need to rethink legal infrastructure to implement a wide range of existing and evolving requirements for safe AI systems that integrate well into our complex economic and social systems.

This infrastructure evolved for human actors and corporations over time, but it is still missing for AI actors and entities. We could start by creating registration schemes for powerful systems—giving governments critically missing visibility into how this transformative technology is evolving and what is known inside private technology companies about its behavior—and identification/registration schemes for AI agents. These are key requirements to ensure accountability under conventional rules governing the marketplace, such as contract enforcement and responsibility for unsafe or unreasonable behavior.

We also need to explore creating durable ID and registration schemes for rapidly evolving autonomous AI agents to ensure that the accountability systems that undergird our complex economies remain intact.  

I'm also working with colleagues through the International Dialogues on AI Safety to develop this legal and regulatory infrastructure at the global level.

Selected Policy Publications

2025
2024

Managing extreme AI risks amid rapid progress

Science
with Yoshua Bengio, Geoffrey Hinton, Stuart Russell, and others
2024

Regulating advanced artificial agents

Science
with Michael K. Cohen, Noam Kolt, Yoshua Bengio, Stuart Russell