Join us for an ERIM Job Market seminar
Abstract
Governments are rapidly adopting artificial intelligence (AI) regulations to balance societal risks against potential productivity gains. Using a difference-in-differences design that exploits the staggered introduction of AI-related bills across 28 U.S. states, I find that proposed regulation reduces the annual probability that an AI startup secures funding and increases the likelihood of acquisition. The effects are concentrated in rules that restrict which AI systems can be built or deployed (“constraining” rules). By contrast, documentation, auditing, and disclosure requirements (“procedural” rules) dampen these effects, consistent with standardized disclosure reducing information frictions around opaque AI technologies. Hiring patterns show that constraining rules increase demand for AI governance roles and shift technical effort from frontier AI research toward deployment of existing systems, while procedural rules have the opposite effect. Startups also reallocate AI-related jobs to non-regulated states.
