US AI Policy Portal — Case study

The Problem

Most AI law trackers answer a basic question:

What AI related laws exist?

I wanted to answer a different one:

What are states regulating, and how are they doing it?

Existing trackers catalog laws by state, but AI laws rarely do just one thing. A single bill may regulate deepfakes, create disclosure requirements, establish penalties, and create an oversight body.

Treating an entire law as a single record makes meaningful comparison difficult.

To compare states structurally, I needed a different unit of analysis.

The Approach

I reviewed enacted AI legislation across all 50 states and DC and converted each law into its underlying provisions.

Using Python and the Claude API, I built a workflow that decomposes legislation into individual policy interventions and classifies each provision across six analytical dimensions:

AI Use Domain Risk Category AI Function Governance Concept Regulatory Layer Legislative Approach

The Claude API generated the first-pass classification of each provision. I then reviewed every provision against the original bill text and corrected the classification where needed.

The result is a dataset designed for analysis rather than cataloging, making it possible to compare how states regulate AI across jurisdictions, sectors, and governance approaches.

What the Tool Does

Browse enacted AI legislation across all 50 states and DC
Compare states side-by-side
Filter provisions across six analytical dimensions
Export the dataset for independent analysis, with links to primary sources

Key Findings

207

State AI laws analyzed

332

Comparable provisions extracted and classified

75%

Of provisions address specific harms such as deepfakes or fraud rather than broader AI governance frameworks

The broader pattern is that AI regulation in the United States is being shaped at the state level. There is no national AI law. Most states are regulating discrete harms, while a smaller group is experimenting with more comprehensive governance frameworks.

By shifting from laws to provisions as the unit of analysis, the Portal makes those differences visible in a way traditional legislative trackers cannot.

Who It's For

Policy researchers tracking AI governance trends
Compliance and legal teams evaluating multi-state requirements
Organizations deploying AI systems across jurisdictions

Methodology

Legislation was sourced from state legislatures, MultiState, and Orrick tracking resources.

Each provision is classified across six analytical dimensions designed to make heterogeneous legislation structurally comparable. The taxonomy includes explicit decision rules for ambiguous cases and is continuously refined as new legislation emerges.

All classifications were reviewed against source documents before publication.