State-level AI legislation data and analytics across all 50 states and territories.
Loading data…
Click a state to browse its bills.
Select a state to load bills.
| State | Bill | Session | Title & Concepts | Core | Adjacent | NCSL | Status |
|---|
Core AI bills (primary subject is AI), Adjacent AI bills (AI as a policy tool in another domain), and NCSL-confirmed bills (advanced in the legislative process) shown together for direct comparison.
Most active states in the most recent legislative session. Core bars (crimson) capture bills where AI is the primary subject; adjacent bars (gold) capture bills where AI appears as a tool within another policy domain. Together they reflect both legislative focus and broader AI policy engagement.
Concepts matched in bills classified as Core AI. The catch-all "adjacent_tech" label is excluded; it marks adjacent-tier bills rather than a specific topic.
The U.S. State AI Policy Tracker is a research tool developed by the Center for Analytics and Innovation with Data (CAID) at the University of Denver. It monitors AI-related legislation introduced across all 50 states, the District of Columbia, and U.S. territories, and cross-references those bills against the National Conference of State Legislatures (NCSL) AI legislation database. The tracker is intended for researchers, journalists, and policy practitioners who need a comprehensive, up-to-date view of where state AI policy is moving.
Legislative data is sourced from Plural Policy (formerly OpenStates), a nonpartisan organization that aggregates bill text, status, and metadata from all 50 state legislatures. We download bulk session files (structured JSON archives) for every available legislative session, resulting in more than 1.4 million collected bills.
Each session file contains bill identifiers, titles, full text where available, sponsor information, committee assignments, and action history. Raw files are stored locally and never modified after download.
Raw session files are parsed and loaded into a local DuckDB database. We use DuckDB because it is a fast analytical database that runs as a single file, no server required. The bills database is treated as append-only: once a bill is written, it is never updated or deleted. Re-running ingestion on the same session replaces only the records for that session.
Each bill record includes the bill identifier, title, full text (where available), session, state, and last-updated timestamp from Plural Policy.
Bills are scanned against a curated keyword pattern library to identify AI-related legislation. Matches are assigned to one of two tiers:
Bills whose primary subject is artificial intelligence, machine learning, large language models, generative AI, algorithmic decision-making, or closely related technologies. These bills are unambiguously about AI — they name it by term or by direct technical equivalent (e.g., neural network, deepfake, automated decision system).
Bills where AI appears as a tool or concern within another policy domain — for example, a healthcare bill that regulates AI diagnostic tools, or an employment bill that addresses algorithmic hiring. These bills do not always use the phrase "artificial intelligence" directly, but they engage with AI-specific technical vocabulary (e.g., natural language processing, computer vision, reinforcement learning) in a policy context.
Matching is applied to bill titles and, where available, full bill text. The pattern library contains approximately 128 patterns across both tiers and is reviewed periodically to improve coverage and reduce false positives. Each matched bill records which specific concepts triggered the match.
A note on the adjacent AI tier: Defining "meaningful" AI relevance is difficult. The adjacent tier is designed to surface bills where AI is substantively addressed, not merely mentioned in passing. To reduce false positives, we validated the pattern library against a hand-coded sample of ~650 bills and removed 36 patterns that were too generic for reliable use in full bill text, including single-word terms like data, bias, token, and temperature that frequently appear in unrelated legislation. The remaining adjacent-tier patterns require AI-specific technical vocabulary to trigger a match.
Flagged bills are cross-referenced against the National Conference of State Legislatures (NCSL) AI legislation tracking, drawn from five sources:
Together these sources provide approximately 2,600 unique state-bill combinations spanning 2019 to the present. Matching is performed by state and bill identifier, with normalization to handle common formatting differences (e.g., "SB 123" vs "S123"). Bills found in any NCSL source are marked NCSL matched in this tracker.
The Bill Browser displays legislative status information from two sources:
A standardized, cross-state bill stage classification (Introduced, In Committee, Passed Chamber, Enacted, etc.) is not currently available from Plural's bulk data. The NCSL status field is the best available proxy for bills that have advanced in the legislative process.
If you use this tracker in published research, please cite it as:
Center for Analytics and Innovation with Data (CAID), University of Denver. U.S. State AI Policy Tracker. https://du-caid.github.io/tracker/
For questions about methodology, data access, or collaboration, contact the CAID lab via the contact page.