Racing to Safety: Tax Policy for AI Safety-By-Design

Abstract

The White House recently announced its vision of artificial intelligence (AI) policy: AI development is a race and America must win it. To that end, a new America’s AI Action Plan directs federal agencies and states to remove regulatory barriers to AI development and accelerate innovation. This approach leaves limited room for regulatory measures that would address the safety risks of powerful AI systems: their behavior in novel domains remains unpredictable, their decision-making opaqueness, and their alignment with human values is uncertain. While experts warn of large-scale accidents, policymakers find themselves in a bind: Regulate AI and cede ground to China, or race ahead and invite disaster? This Article examines this apparent dilemma and uncovers a stark market failure at its heart. AI labs currently invest tens to hundreds of times more in making their systems powerful than in making them safe. This imbalance persists because labs capture immense rewards from powerful AI while socializing the gravest tail risks—systemic failures, malicious exploitation, and loss of human control. The traditional regulatory tools that would force labs to internalize these risks are largely unavailable because they are seen as hobbling U.S. firms exclusively. We argue that the tax code—an instrument largely absent from AI safety debates—offers an unexpected resolution. This Article advances a new understanding of how tax policy can govern emerging technologies under conditions of geopolitical competition. Building on innovation pluralism theory and historical precedents, we develop a comprehensive framework that harnesses the tax code to embed safety-by-design principles into AI development. This approach exploits a critical insight: Tax incentives can make safety research profitable without making capability research unprofitable. The framework leverages three mechanisms: a unique credit for AI safety research, consumer incentives for secure AI products, and redistribution mechanisms that fund public safety initiatives by recapturing tax benefits from unsafe developers. By aligning private incentives with public welfare, the tax code ensures that winning the AI race means deploying systems that are not merely powerful but demonstrably secure. In the high-stakes competition for AI supremacy, safety becomes not a burden but America’s decisive advantage.