A Roadmap for Governing AI

In the AI Story

The paper emerged from Allen's work with the GETTING-Plurality research network at Harvard and reflects sustained engagement with technologists, policymakers, and democratic theorists across the preceding four years. Its seventeen recommendations are organized across multiple levels of governance—federal, state, local, and transnational—reflecting Allen's insistence that AI governance requires a multi-layered institutional architecture no single level can provide alone.

The federal-level proposals include licensing requirements for firms developing frontier AI models, analogous to licensing regimes for financial institutions and nuclear facilities. The logic is that organizations developing technology with systemic implications should face regulatory scrutiny commensurate with the power they wield. The proposal is controversial—it would subject technology companies to a degree of federal oversight they have historically resisted—but Allen argues that the alternative is governance by unaccountable private actors whose decisions shape public life without public accountability.

The state-level proposals include the establishment of AI offices in state governments to enhance accountability and enable responsive governance. The logic is that state governments are closer to the specific impacts of AI deployment—on education, labor markets, public services, local economies—than federal institutions can be, and that effective governance requires institutional capacity at the level where impacts are felt. The proposal draws on the historical precedent of state-level responses to earlier technological transitions, including labor regulation during industrialization and environmental regulation in the twentieth century.

The paper also addresses what Allen calls public goods and democratic infrastructure—the institutions required to ensure that AI's benefits are broadly distributed and that its development serves democratic purposes rather than merely commercial ones. These include public investment in AI infrastructure as an alternative to monopolistic private control, support for open-source AI development, educational reforms to develop participatory readiness, and transitional support for workers whose roles are being restructured by AI deployment.

Origin

'A Roadmap for Governing AI' was published in 2025 through Allen's work at the Harvard Kennedy School and the GETTING-Plurality research network. The paper builds on 'How AI Fails Us' (2021) and applies the framework of Justice by Means of Democracy (2023) to the specific institutional challenges of AI governance.

Key Ideas

Seventeen recommendations. The paper presents specific, actionable proposals rather than abstract principles.

Multi-layered architecture. AI governance requires coordinated institutions at federal, state, local, and transnational levels.

Federal licensing. Organizations developing frontier AI should face regulatory scrutiny commensurate with the power they wield.

State-level capacity. AI offices in state governments are needed to address impacts at the level where they are felt.

Public goods investment. Democratic governance of AI requires public investment in infrastructure, education, and transitional support.

Debates & Critiques

Critics from the technology industry argue that the licensing and regulatory proposals would slow innovation, disadvantage American firms relative to international competitors, and concentrate regulatory capture risks. Allen's response is that unregulated AI development is already concentrating power in ways that threaten democratic governance, and that the choice is not between regulation and innovation but between democratic regulation and private capture of regulatory outcomes.