The EU AI Act vs US regulation
One statute against a patchwork: how the two regimes actually differ on scope, enforcement, penalties, and dates, cited to the register.
The comparison most compliance teams need is not philosophical. It is operational: the EU regulates AI through one extraterritorial statute with an enforcement agency and a fine schedule, while the United States regulates it through eleven state laws, sector agencies, and private lawsuits, with a federal government trying to erase the state layer in court. Since August 2, 2026, both sides of the Atlantic impose transparency duties on the same day, by design.
| European Union | United States | |
|---|---|---|
| Architecture | One horizontal statute, Regulation 2024/1689, phased 2025 to 2028 | No federal statute; eleven state regimes plus sector agencies |
| Who enforces | AI Office plus national market-surveillance authorities | State attorneys general, FTC, EEOC, and private plaintiffs under BIPA |
| Risk model | Bans, high-risk tiers, and transparency duties by category | Use-based bans (Texas), disclosure duties (California), audits (NYC, Illinois) |
| Transparency | Article 50 in application since August 2, 2026 | California's Transparency Act operative the same day; no federal duty |
| Frontier models | GPAI obligations since 2025; Commission fining power since August 2026 | State template: California, New York, Illinois; 10^26 ops, $500M revenue |
| Top exposure | €35M or 7% of turnover; €15M or 3% for transparency | BIPA statutory damages stacking per scan; TRAIGA capped at $200K per violation |
| Reach | Extraterritorial: any output used in the EU | State laws reach activity touching their residents |
| Next dates | December 2, 2026 marking grace ends; December 2027 high-risk | January 1, 2027: Colorado, New York, Illinois all bite |
Does the EU AI Act apply to US companies?
Yes, whenever their systems or outputs are used in the EU, regardless of where the company sits. A US provider whose model reaches European users carries the same transparency, GPAI, and, from December 2027, high-risk obligations as a European one. That extraterritorial reach is why most global products build to Brussels and subtract.
Is there a US equivalent of the EU AI Act?
No, and none is coming soon. Congress has passed nothing comprehensive, the executive branch is trying to suppress state laws rather than replace them, and the actual American rulebook is eleven state regimes layered over sector agencies. The closest structural echo is the three-state frontier template in California, New York, and Illinois.
Which is stricter, the EU or the US?
The EU is broader; parts of the US bite harder. Nothing in America matches the AI Act's scope or its 7 percent turnover ceiling. But Illinois' BIPA produces larger real-world payouts than any EU enforcement to date, because statutory damages stack per violation and private plaintiffs, not regulators, bring the cases.
What happened on August 2, 2026?
The first synchronized compliance date across the Atlantic. The EU's Article 50 transparency duties went into application and the Commission's enforcement powers over general-purpose models became applicable, while California's AI Transparency Act reached its operative date, aligned with the EU deadline by design.
The full detail lives on the EU entry, the US federal entry, the state tracker, and the penalty comparison.