Congress is finally putting something on paper.
Last week, Representatives Lori Trahan and Jay Obernolte released a 269-page discussion draft of the Great American AI Act, the first serious attempt at a federal framework for governing AI in the US. It's bipartisan, it's detailed, and it's a different animal from the toothless 2023 bill of the same name that just asked the Commerce Department to study the issue.
The draft is organized into four parts:
- Frontier models: Companies building the largest systems would face mandatory documentation, 72-hour incident reporting, and independent safety verification.
- Workforce: New rules around labor market impact, training programs, and disclosure when AI is used in hiring or firing.
- Federal use: Stricter procurement rules for any AI system the government touches, building on a narrower Senate contractor bill that already passed earlier this month.
- Innovation: Tax credits and research funding aimed at keeping American AI competitive with Chinese labs.
The Senate's contractor bill is the warm-up act. The House framework would reach into almost every commercial use of AI in the country, which is why industry is paying close attention. And the most controversial part of the bill has nothing to do with safety. It's the preemption clause.
The draft would block states from writing their own AI laws, which is exactly what tech companies have been lobbying for. Neil Bradley, chief policy officer at the U.S. Chamber of Commerce, said the preemption provision "alone is worth every dollar we've spent on this process," arguing American companies can't operate while juggling seventeen different state definitions of an AI system. Trahan framed her own pitch around national security, calling AI threats to the workforce and public safety "here and growing by the day." Representative Erin Houchin, a Republican co-sponsor, put the preemption argument more bluntly: a "patchwork of fifty different state laws" would just hand the lead to China.
Not everyone is sold. Senator Ted Cruz, who voted against the narrower Senate bill, argued that "federal paperwork cannot substitute for market competition among AI vendors." His real worry is compliance cost on startups. Consumer advocates have a different complaint, pointing to an open-source audit exemption they've called "a loophole big enough to drive a frontier model through." And state lawmakers, especially in Massachusetts and California, are organizing against losing their seat at the table entirely.
Senator John Curtis, a Republican, summed up the shift in an interview with Politico: "I think we're landing more and more in a place where everybody's realizing you need some type of government oversight. I think we're still struggling with what that is."
Legal analyses from DLA Piper, ArentFox Schiff, and Akin Gump all reach the same read: even in discussion-draft form, this is the most substantive piece of AI legislation Congress has produced. Tech Policy Press goes further, calling it a likely template for whatever final bill emerges.
For three years the running joke about US AI policy has been that Washington was waiting for someone to write the bill. Now someone has, and the fight shifts from whether Congress regulates AI to whose version of regulation wins. The preemption clause is where this gets decided. If it survives, fifty state legislatures lose their voice on AI, which is what Silicon Valley wants and what consumer groups will spend the next six months trying to stop. Europe just gutted its own AI Act this month, so the window for the US to write rules that other countries actually copy is narrower than it looks.
