Legislating in the Shadow of Congress: Selective Convergence in Federal-State AI Bill Language

• In Event: Federalism and Policymaking

Abstract

How does federal legislative language shape state-level policymaking? As artificial intelligence rapidly permeates consequential domains, governments at all levels face mounting pressure to establish regulatory frameworks. In the United States, this has produced a fragmented legislative landscape in which states have moved aggressively to fill a federal vacuum, introducing thousands of AI-related bills even as comprehensive federal legislation remains elusive. Yet despite this apparent decentralization, it remains unclear whether state legislators are developing independent regulatory vocabularies or drawing systematically on federal legislative templates. Understanding this dynamic is essential not only for mapping the architecture of emerging AI governance, but for evaluating how legislative language itself functions as a mechanism of policy coordination across levels of government. This study examines textual convergence between federal and state artificial intelligence legislation in the United States using an original dyadic corpus of 230 federal and 1,633 state AI-related bills introduced between 2019 and 2025. Applying three complementary text similarity measures to over 300,000 federal-state bill pairs, the analysis finds that convergence is selective rather than pervasive, concentrated in a small number of policy domains including algorithmic accountability, synthetic media regulation, and AI literacy. When convergence does occur, it tends to happen rapidly following federal bill introduction rather than through gradual diffusion over time. Contrary to expectations from the partisan diffusion literature, partisan alignment between bill sponsors does not predict textual convergence, suggesting that language borrowing in emerging technology governance is driven by the availability of credible federal templates rather than by partisan networks. These findings extend the text-as-policy approach to the vertical dimension of American federalism and suggest that federal legislative proposals, even when unenacted, exert meaningful influence on the vocabulary of state-level AI regulation. These findings carry several implications for AI governance and legislative practice. First, they suggest that federal inaction does not preclude federal influence. Even unenacted bills can still structure the terms of state-level deliberation by supplying authoritative regulatory language that state legislators selectively adopt. This has direct consequences for how federal policymakers understand their role: even proposals that fail to advance through Congress may shape the trajectory of subnational governance in consequential ways. Second, the domain-specificity of convergence points to areas such as algorithmic accountability, synthetic media, AI literacy where federal-state coordination is already occurring organically, and where targeted federal guidance or model legislation could accelerate coherent, nationally consistent regulatory development. Third, the absence of a partisan convergence effect suggests that AI governance may represent a domain where technical complexity and the need for credible templates override partisan signaling, offering a potential opening for bipartisan legislative coordination at both levels. For policymakers and advocates seeking to shape the emerging AI regulatory landscape, these findings underscore the strategic value of investing in high-quality, clearly drafted federal proposals regardless of their immediate prospects for enactment.

Author

• Eunji Emily Kim, University of California San Diego
  Presenting Author