From Symbolic Attention to Operational Rules: Explaining Variation in State AI Legislation

• In Event: AI Governance and Regulation: From State to Federal

Abstract

State legislatures have become central venues for AI policymaking in the United States. Since 2019, lawmakers have introduced and enacted measures addressing algorithmic discrimination, government use of automated systems, election-related synthetic media, consumer transparency, educational technology, and health-related AI applications. However, legislative activity has differed sharply across states not only in volume, but in design. Some states have enacted operational rules with clear obligations and enforcement mechanisms, while others have relied on task forces, study commissions, pilot authorizations, or narrowly targeted prohibitions. The variation suggests that the politics of AI legislation cannot be captured by simple counts of bills introduced or enacted. This study explains differences in the timing, content, and legal strength of state AI legislation between 2019 and 2026. The analysis draws on an original dataset of AI-related state bills and enacted laws coded by policy domain, target sector, governing instrument, enforcement mechanism, and level of legal obligation. The empirical sample includes high-activity states in which AI became a visible legislative agenda, together with a comparison group of lower-activity states. Legislative records are linked to measures of partisan control, legislative professionalism, committee capacity, technology-sector density, prior digital governance experience, and participation in interstate policy networks. The empirical strategy combines event-history analysis with comparative policy design analysis. The first component models the timing of initial major AI enactments and evaluates whether legislative uptake is associated with geographic proximity, partisan similarity, or policy-network diffusion. The second component examines legislative content, distinguishing among disclosure regimes, anti-discrimination safeguards, public-sector use rules, sectoral regulation, innovation-promoting measures, and commission-based responses. Comparative case analysis is then used to trace why states facing similar technological pressures nevertheless adopted different legal instruments and enforcement structures. The analysis yields three preliminary conclusions. First, partisan control is associated with legislative activity, but it does not fully account for variation in legal design. States with stronger legislative infrastructure and prior experience governing digital systems are more likely to produce operational rules rather than symbolic or temporary measures. Second, policy diffusion appears selective and mediated by institutional capacity: visible state models matter, but they are adapted rather than copied wholesale. Third, legislative volume is a poor proxy for governance depth. Some states generate extensive legislative activity without building enforceable regulatory institutions, whereas others enact fewer measures but produce more consequential legal change. The study contributes to research on technology governance by distinguishing agenda-setting from durable institutionalization. More broadly, it demonstrates that variation in state AI legislation reflects differences in policy design capacity as much as differences in political preference, offering a more precise framework for understanding subnational regulation of emerging technologies.

Author

• Ruoxi Li, The Hong Kong University of Science and Technology (HKUST)
  Presenting Author