This study was originally published by the American Consumer Institute (ACI).
Read the comprehensive study here.
Executive Summary
Artificial intelligence (AI), an emergent technology still in its infancy, holds immense promise to improve American living conditions, expand consumer choice, and promote general economic flourishing. Rather than embrace this novel technology, state and local governments have responded with a regulatory briar patch of fear-based laws and precautionary policies that threaten to undermine this rapidly improving industry before it has a chance to be widely developed and utilized.
To highlight the worst offenders in state-level AI governance, this paper (1) introduces the many problems with the growing AI patchwork; (2) suggests a national policy framework as the ideal solution; (3) identifies and categorizes recent state legislative proposals; and (4) considers alternative legislative frameworks that offer a better approach to some of the more troubling laws currently pending or already passed. In evaluating these state-level laws, this paper introduces the AI Terrible Ten: ranking the worst state regulatory ideas and specific laws in each category of AI governance from 2025. The AI Terrible Ten include:
- "AI Fairness” & Anti-Bias Laws: Colorado’s Consumer Protections for Artificial Intelligence Act (signed by Governor)
- Precautionary Principle-Based Regulatory Frameworks: Hawaii Artificial Intelligence Safety and Regulation Act
- Algorithmic Pricing Regulations: New York’s Algorithmic Pricing Disclosure Act (signed by Governor) & California’s Amendments to the Cartwright Act (signed by Governor)
- Automation Taxes: New York's Robot Tax Act
- Omnibus AI Packages: Florida’s Artificial Intelligence Bill of Rights
- AI Mental Healthcare Bans: Nevada AB 406 (signed by Governor)
- Frontier Model Safety Laws: New York’s Responsible AI Safety and Education Act (RAISE) (signed by Governor)
- Content Provenance and AI Transparency: California’s AI Transparency Act Amendments (signed by Governor)
- Output and Model Ownership: Arkansas’ An Act Regarding the Ownership of Model Training and Content Generated by a Generative Artificial Intelligence Tool (signed by Governor)
- Criminal Conduct Escalators: Michigan's Amendments to 1931 PA 328
As the AI Terrible Ten makes clear, the United States finds itself at a technological crossroads: embrace technological innovation or subject the technology to mountains of paperwork, tickytack compliance regiments, unclear reporting requirements, distortionary tax policies, readily abusable state government statutes, and even outright prohibitions on many applications of the technology’s most promising use-cases. Consumers benefit more from pro-freedom AI policies than heavy-handed restrictions that undermine life-enriching innovations.
Although the overwhelming trend is toward micromanagement and over-regulation, some states—Montana and Utah—have instead embraced future-forward governance and novel approaches to regulatory co-learning in the intelligence era. In 2026, more states will face a choice between freedom and control. This paper identifies why repression should not be the first order of business and why freedom and opportunity should be the default. We outline proinnovation, light-touch AI governance options that can strike a more sensible balance than the heavy-handed policies currently being considered.