The Great American AI Act Is a Real Bill. It Is Not Going to Pass. Here's What 17% Actually Means.

First, the definitional housekeeping that the original version of this forecast skipped. When we forecast 'US Congress passes federal AI legislation,' we do not mean any bill with 'AI' in the title. The US Congress has already passed AI-relevant legislation — the AI in Government Act, CREATE AI Act provisions in the 2024 NDAA, and various AI-adjacent authorizations. What we are forecasting is whether Congress passes a comprehensive standalone bill that imposes mandatory obligations on AI developers: safety frameworks, incident reporting requirements, whistleblower protections, preemption of state law. The Great American AI Act of 2026 is precisely that. It is the resolution target. Prior bills are not.

Now, on the evidence. The June 4 Obernolte-Trahan discussion draft matters because bipartisan sponsorship on a 269-page framework is not nothing — it is evidence that at least two members of Congress believe there is a coalition to be built around federal preemption as the deal-making mechanism. The theory of the bill is essentially: industry trades mandatory federal obligations for three years of state law preemption. That is a real offer that real lobbyists will take seriously. Colorado's AI Act, which we now have confirmed takes effect June 30 after an industry-pressured delay from the original February 2026 date, gives the preemption argument its sharpest edge yet. A June 30 Colorado deadline creates a live, ticking compliance burden for firms operating in that state — and Colorado is explicitly the kind of state-level law the Great American AI Act would preempt. This is the strongest proximate signal for the upside scenario.

But proximate is not direct, and this is where we need to be careful about the inference chain. The June 2 Trump AI cybersecurity executive order — which reduced pre-release model review from 90 days to 30 days and explicitly rejected mandatory licensing or preclearance — is not direct evidence of how the White House would respond to the Great American AI Act's mandatory provisions. It is suggestive evidence. The EO signals a preference for voluntary engagement over mandatory frameworks, which cuts against the bill's core mechanism. But EOs on cybersecurity and comprehensive frontier governance legislation are different instruments, and it is possible the administration calculates that federal preemption of state law is worth accepting some mandatory provisions. We do not know that. We are treating the EO as a strong signal of White House resistance, not a certainty.

Here is the decomposition behind 17%. Roughly 8% reflects the preemption-deal scenario: industry flips hard for federal preemption as Colorado-style laws multiply, lobbying pressure shifts from 'no bill' to 'this specific bill,' and a stripped-down version focused on preemption with lighter mandatory provisions clears both chambers before the end of 2026. Roughly 5% reflects a post-election lame-duck window if the November 2026 midterms produce a political environment where both parties have incentive to lock in a framework before new Congress. Roughly 4% is residual probability on scenarios we have not modeled well — a major AI safety incident that breaks the political logjam, or international pressure from the EU August enforcement date creating an unexpected urgency. That adds to 17%, and it is honest about how thin the path is. The single biggest thing keeping us from going below 10%: we think markets are underpricing the political pain of a fragmented multi-state compliance environment. If five more states pass Colorado-style laws in Q3, the preemption argument becomes much harder for Republican members to resist. What would move us above 25%: committee markup of the Great American AI Act before the August recess. That would be direct evidence of legislative momentum, not just discussion-draft posturing. No markup, and we drop below 12%.