Bipartisan 'Great American AI Act' Draft Proposes Three-Year Preemption of State AI Laws

On 4 June 2026 Representatives Jay Obernolte (R-Calif.) and Lori Trahan (D-Mass.), joined by four co-sponsors from both parties, released a discussion draft of the Great American AI Act. The text is the most concrete congressional attempt so far to answer a question that has divided the industry and state capitals for two years: whether the United States should have one federal rulebook for frontier AI development or continue with a patchwork of state statutes.

The draft's central mechanism is a three-year preemption of state laws and regulations that specifically govern how AI models are developed. In exchange it would establish federal safety standards for developers, with state attorneys general retaining authority to enforce those federal standards. The preemption is deliberately narrow: laws addressing the use or deployment of AI, such as the companion-chatbot statutes now enacted in 14 states, health-insurer restrictions and algorithmic pricing limits, would be untouched. A sunset clause ends the preemption after three years unless Congress acts again.

Reaction split along familiar lines. Safety advocates argued that freezing state developer rules before federal ones are proven is a mistake; Americans for Responsible Innovation's Brad Carson called it a "generational mistake". Some industry groups, including NetChoice, objected from the other side to what they saw as aggressive auditing requirements in the federal standard. The sponsors said they would revise the draft after stakeholder feedback before formal introduction.

The draft arrives against a shifting backdrop. State legislatures had enacted 109 AI-related laws by 1 July 2026, slightly behind 2025's pace, and analysts attribute the slowdown in part to expectations of federal preemption following the White House's December 2025 push to centralise AI policy. Illinois was simultaneously advancing a frontier-model bill requiring third-party audits, exactly the kind of law the draft would suspend. In Europe, by contrast, the same month saw the AI Act's high-risk obligations delayed rather than displaced, so multinational firms face divergence in both timing and structure.

Whatever its fate, the draft signals that developer-level obligations, including safety testing, incident reporting and audits, are now on the table in Washington with bipartisan sponsorship. Enterprises that only deploy AI are less directly affected, but the line between developing and deploying is blurry for any organisation that fine-tunes models or builds agents.

What it means for leaders

• Classify your role. Determine whether fine-tuning, distillation or agent construction could make your organisation a "developer" under the draft's definitions; the answer changes which regime applies.
• Do not pause state compliance. Deployment-focused laws (Colorado, California, Texas, New York, plus the companion-chatbot wave) remain in force under every version of the proposal; keep impact assessments and notice obligations on schedule.
• Build to a common denominator. A governance programme anchored in NIST AI RMF and ISO/IEC 42001 maps to both the likely federal standard and to EU obligations, reducing rework as the legal landscape moves.
• Prepare for audits regardless. Both the House draft and the Illinois bill contemplate third-party assurance; establishing documentation, evaluation records and model cards now avoids a scramble later.
• Engage in the comment window. Discussion drafts are revised on the basis of stakeholder input; sector associations and large deployers should submit specific, evidence-based feedback on definitions and enforcement.