The United States now has more AI legislation in motion than any other jurisdiction in the world, and less coherence than almost any of them. A May 15 analysis in Fortune, co-authored by Yale professor Jeffrey Sonnenfeld and cognitive scientist Gary Marcus, documents the result of that paradox: over 1,200 AI-related bills were introduced across state legislatures in 2025 alone, with nearly 150 enacted — each pursuing a different theory of what AI risk looks like and who is responsible for managing it. California emphasizes developer transparency. New York mandates incident reporting. Texas created a 36-month regulatory sandbox. Connecticut focuses on consumer disclosure. None of these frameworks speak to each other.
The federal picture is not cleaner. A December 2025 executive order tasked agencies to build a "national policy framework for AI" while preempting state regulation through federal lawsuits and conditional funding — but established no actual standards or compliance requirements. The result, as legal analysts at Gunderson Dettmer noted in their 2026 AI Laws Update, is that U.S. companies face a patchwork of binding state obligations with no clear federal floor to build compliance programs around. Colorado's AI Act took effect in February 2026, with enforcement beginning June 30. The EU AI Act becomes fully applicable August 2, 2026. Both deadlines are imminent; most mid-market companies are not ready for either.
Sonnenfeld and Marcus propose that legislators adopt a three-stage test before passing any AI bill: whether existing law already addresses the harm, whether the cost-benefit case for new regulation is sound, and whether the law is durable and enforceable as the technology evolves. The proposal reflects a deeper frustration: the policy community has produced volume, not clarity. As the EU prepares to begin enforcement with a working compliance architecture and China advances a state-coordinated AI strategy, the U.S. is, in the words of the Fortune authors, operating without "a shared test to determine whether legislative efforts constitute good policy." The August 2026 EU AI Act activation date may serve as the forcing function that finally pressures Washington toward a substantive response.