How is AI Regulation: Bills vs Laws constructed?
Published by Deepstory Research as the methodology attached to AI Regulation: Bills vs Laws. Scope and source attribution are bounded by that Evidence Brief.
Procedure and scope
Two counts that are routinely conflated. BILLS INTRODUCED measures legislative attention; LAWS ENACTED measures actual compliance burden. Roughly 1,200 AI bills were introduced in 2025 and 145 became law — a conversion rate near 12%. Coverage that cites introduction counts as evidence of a regulatory wave overstates the burden by close to an order of magnitude. The two charts are NOT a matched pair: the 2024 introduction count and the 2026 enactment count were not retrieved, so each chart carries the years its source publishes. Do not compute a conversion rate by reading one bar against the other for a year that appears on only one chart. CAVEAT: the 2026 introduction figure is a PARTIAL YEAR (to March 2026, 45 states) and is not comparable to the full-year 2025 bar beside it. Scope note: enactment counts include narrow single-issue statutes. The 2024 cohort was dominated by nonconsensual explicit deepfake laws, passed in 22 states — counting those alongside comprehensive AI-governance frameworks flatters the total. By July 2026 roughly five states had enacted comprehensive frameworks for private-sector AI in high-stakes decisions. For context beyond the charts: the EU AI Act entered into force in 2024, GPAI obligations applied from 2 August 2025, the bulk of the regulation and its enforcement rules apply from 2 August 2026, and high-risk system obligations follow from 2 December 2027 (standalone) and 2 August 2028 (embedded in products). Those are dates rather than measurements, so they are not charted. Re-verified 2026-08-11.
Known failure modes and limits
No reviewed failure-mode list is published.