The Great American AI Act: What US Federal AI Regulation Means for Every Business

This week, US Representatives Jay Obernolte and Lori Trahan unveiled a 269-page discussion draft of the Great American Artificial Intelligence Act. Its headline provision: a three-year federal preemption of all state AI laws governing frontier model development.

If passed, it would instantly nullify more than 600 state-level AI bills — from California's SB 1047 successors to New York's emerging model liability frameworks — and replace them with a single federal standard. It is the most significant AI governance development in the United States since the Biden administration's 2023 executive order on AI safety, and it arrives at a moment when both the EU AI Act's enforcement provisions are coming into full force and global businesses are trying to make sense of a fragmented, increasingly costly compliance landscape.

Understanding what this bill actually proposes — and what it does not — matters well beyond US borders.

What the Bill Actually Proposes

The draft legislation has three core pillars.

A three-year federal preemption. For frontier AI model developers — companies training or deploying AI systems above defined capability thresholds — the bill would suspend state-level AI liability and governance requirements for thirty-six months. The explicit purpose is to prevent a patchwork of conflicting state mandates from fragmenting the US AI industry during what the bill's sponsors describe as a "critical development window."

A light-touch federal floor. Rather than imposing detailed prescriptive requirements, the federal standard establishes baseline obligations: safety testing for frontier models before deployment, incident reporting to a designated federal body, and mandatory disclosure requirements for AI-generated content in specific high-stakes domains (elections, healthcare, legal). The framework is deliberately designed to be less burdensome than the EU AI Act.

A pre-emption carveout for states on harm. States retain the right to regulate AI harms through existing tort law and consumer protection frameworks. What they cannot do, under the proposed bill, is create new AI-specific licensing requirements, capability restrictions, or liability regimes for frontier model developers during the preemption window.

The bill is currently a discussion draft — not yet introduced for a floor vote. But discussion drafts at this detail level (269 pages, developed with significant industry consultation) typically reflect the shape of eventual legislation. The three-year window gives Congress time to pass something more permanent before the preemption lapses.

Why the US Chose Federal Preemption

To understand why this bill exists, you need to understand what the alternative looks like.

By mid-2026, 46 US states have passed or are actively considering AI-related legislation. The bills vary enormously in scope and approach. California has focused on developer liability for model outputs. Texas has prioritized algorithmic transparency in government use. Colorado and Connecticut have enacted employment-related AI disclosure laws. Illinois has focused on biometric data and facial recognition.

For any company developing AI systems and deploying them across state lines — which is essentially every enterprise AI vendor — this means maintaining 46 parallel compliance programs. The compliance cost is not theoretical. Legal teams at major AI vendors estimated the collective cost of the California-era bills alone at over $800 million annually.

Federal preemption is the industry's answer: one standard, one compliance program, three years of regulatory stability during a period of intense model development.

The political calculus is straightforward. The bill's sponsors — Obernolte (Republican) and Trahan (Democrat) — have framed it explicitly in terms of US competitiveness against China. The argument that a fragmented state regulatory landscape hands a structural advantage to Chinese AI development has proved persuasive across party lines. Whether it survives the legislative process intact is another question.

How This Compares to the EU AI Act

The contrast between the emerging US approach and the EU AI Act is instructive for any business operating in both markets.

The EU AI Act, which began full enforcement in 2026, is a prescriptive, risk-tiered framework. High-risk AI systems — in employment, healthcare, critical infrastructure, and a growing list of other domains — require conformity assessments, technical documentation, human oversight mechanisms, and registration in an EU database before deployment. Frontier model providers face additional obligations under the General Purpose AI provisions: capability evaluations, adversarial red-teaming, and transparency reporting.

The Great American AI Act, by contrast, is a floor-setting, innovation-first approach. It establishes baseline obligations but explicitly does not require pre-deployment approvals, conformity assessments, or regulatory licensing for AI systems. The philosophy is different: the EU treats AI governance as a product safety problem requiring upfront certification; the US bill treats it as a disclosure and accountability problem requiring transparency and post-hoc liability.

For global AI vendors, this creates a genuine compliance bifurcation. Meeting EU AI Act requirements does not satisfy the US bill's requirements, and vice versa. The EU's conformity documentation does not map cleanly onto the US disclosure and incident reporting framework. Compliance teams are going to need two distinct programs — at minimum.

The silver lining: the US framework is, on most dimensions, less burdensome than the EU one. Companies already compliant with the EU AI Act will find the US baseline achievable. The reverse is not necessarily true.

What This Means for Swiss Businesses

Switzerland's position is, as usual with EU matters, distinctive. Swiss businesses are not directly subject to the EU AI Act under its current scope (Switzerland is not an EU member), but they are practically subject to it if they deploy AI systems to EU customers — which most Swiss enterprises of any scale do.

The Great American AI Act adds a second compliance dimension for any Swiss company operating in the US market or using AI vendors headquartered there.

Three practical implications:

1. Your AI vendor's compliance posture is about to change. US-headquartered AI vendors — OpenAI, Anthropic, Google DeepMind, Microsoft — will adjust their model documentation, incident reporting, and transparency disclosure practices to align with the federal standard once it passes. This may affect the audit documentation they provide to customers, the content filtering configurations they offer, and the incident notification timelines you can expect. If you have compliance SLAs tied to AI vendor behavior, review them now.

2. The EU AI Act remains your primary regulatory reality. For Swiss businesses, the EU AI Act governs more of your practical AI use than any US legislation will. The Great American AI Act's preemption window applies to frontier model developers — which most Swiss SMEs are not. What matters for a Swiss company deploying an AI system in an EU-regulated context is still the risk classification, documentation, and human oversight requirements of the EU AI Act. Nothing in the US bill changes that.

3. Watch the three-year window. The preemption sunset in 2029 is the important date. Congress is expected to use the three-year window to develop permanent federal AI legislation. Whatever emerges from that process will set the global standard for how AI systems are governed in the world's largest AI market. The shape of that legislation — closer to the EU model or further from it — will determine whether global AI compliance converges or permanently bifurcates. For any business planning a multi-year AI investment roadmap, the 2029 regulatory horizon is a planning assumption, not a footnote.

The Broader Signal

What the Great American AI Act signals — regardless of whether it passes in its current form — is that the era of AI governance by executive order and regulatory guidance is ending. Legislatures on both sides of the Atlantic are now setting the rules.

For businesses, this is actually clarifying. The ambiguity of the last three years — where enforcement priorities shifted with each administration and compliance advice had a short shelf life — is giving way to statutory frameworks with defined requirements, defined penalties, and defined timelines.

The compliance burden is real. But predictability has value. A business that builds AI governance infrastructure now — audit trails, model documentation, human oversight mechanisms, vendor compliance reviews — is building something that will hold its shape across regulatory jurisdictions. A business that waits for the rules to finalize will find the landscape more expensive to navigate, not less.

The Great American AI Act is not law yet. But it is the clearest statement yet of where US AI governance is heading. For any business with AI investments of any significance, the time to read the draft is now — not after the final vote.

TecMinds advises Swiss SMEs on AI strategy, compliance, and implementation. If you are working through what the EU AI Act or emerging US AI regulation means for your specific operations, get in touch.

Sources

• Great American Artificial Intelligence Act — Legislative Discussion Draft, June 2026
• AI Legislative Update: June 12, 2026 — Transparency Coalition
• Battle for AI Governance: White House's Plan to Centralize AI Regulation — Vorys
• US AI Regulations 2026: Federal Orders, State Laws — VerifyWise
• Top AI News for June 2026 — AIapps.com