A new bipartisan US proposal would do something Silicon Valley loves to argue about and states hate to surrender: move oversight of the most advanced AI systems from a patchwork of local laws to a federal framework. The draft, called the Great American AI Act, would require developers of frontier models to report regularly to Washington, spell out how they plan to reduce serious risks, and open the door to external audits.
The Great American AI Act could also override some state AI rules through a federal preemption mechanism, setting up a major fight over who should regulate advanced AI in the US. For an industry that already faces a growing list of state-level obligations, a single federal rulebook would be far easier to navigate – unless you are the state attorney general who just wrote the stricter version.
What the Great American AI Act would create
The proposal, introduced by Republican Jay Obernolte and Democrat Lori Trahan, would establish the Center for AI Standards and Innovation, or CAISI, inside the National Institute of Standards and Technology at the US Department of Commerce. Its job would be to assess frontier models over three years, assuming Congress approves the $300 million funding request separately.
- Regular reporting from developers of ”advanced” AI models
- Plans to reduce major risks, including cybersecurity threats
- External audits to check compliance
- CAISI housed within NIST under the Commerce Department
- $300 million in proposed funding, pending congressional approval
The state preemption fight is the real flashpoint
The most contentious part of the bill is not the reporting template or the audit schedule. It is the possibility that federal rules would crowd out existing state laws, including tougher requirements that have already been passed in some places. That is a familiar pattern in US tech policy: federal lawmakers promise consistency, then spend months arguing over whether ”consistency” is just another word for lowering the bar.
Supporters say that is the point. Trahan has argued that AI risks are already showing up in national security, public safety, and labor markets, while Obernolte says the goal is a unified framework shaped with expert and industry input. That sounds sensible enough until you remember that the most powerful companies usually prefer one federal regulator to a mess of state ones, especially when state rules start getting sharp.
Why critics think Big Tech benefits most
Public Citizen and other advocacy groups are already attacking the proposal, saying it ignores harms such as algorithmic discrimination, fraud, teen safety, and deepfakes while weakening protections that states have already put in place. They also argue the bill could tilt the field toward companies like OpenAI, Meta, Google, and Anthropic, because a federal regime may replace stricter local rules rather than layering on top of them.
There is political risk for the bill’s sponsors, too. The source material says polling cited by advocacy groups suggests many voters in the lawmakers’ districts are skeptical of rolling back state AI regulation. That does not kill the bill, but it does explain why its authors are presenting it as a discussion draft rather than a finished product: in Washington, ”open for feedback” is often code for ”we know this is going to get messy.”
What happens before the bill reaches Congress
For now, the Great American AI Act is still in public discussion and could change substantially before formal introduction. If it survives the next round of lobbying, the real question is whether Congress wants a national AI rulebook that smooths compliance for developers, or one that also preserves the tougher experiments already happening in the states. My bet: the preemption fight will decide the bill’s fate long before the technical details do.