The Patchwork Is Real. The Pressure Isn't Enough. We're Holding at 18%.

Let's be precise about what we're forecasting before we argue for it. The 18% applies to a specific window and a specific threshold: standalone legislation or a substantial rider establishing federal AI governance standards — not a narrow export control amendment, not a resolution, not an executive order — signed into law before November 2026. If an AI governance provision were to pass as a rider on the NDAA or an intelligence authorization bill, we would count that as YES. We're stating this explicitly because the most plausible near-term path runs exactly through that vehicle, and readers deserve to know where the resolution line sits before we tell them how close we think we are to crossing it.

The 18% is anchored to a reference class: major sector-specific regulatory legislation in polarized Congresses — think financial reform, telecom overhaul, healthcare — passes at roughly 15-20% when introduced without durable bipartisan co-sponsorship on both chambers' versions simultaneously. That's the base rate. We're sitting just above it because the preemption dynamic is genuinely unusual: the Trump administration wants federal preemption to neutralize state-level regulation, and that creates a narrow alignment with Republican congressional priorities that doesn't exist for most tech regulation. But 'unusual' doesn't mean 'sufficient.' The state bill count that matters here isn't 1,200+ introduced bills — raw introduction volume is a measure of political energy, not regulatory divergence. What creates actual compliance pressure on Congress is enacted state law. New York's RAISE Act joining California's SB 1047-era framework means large AI developers now face materially different incident reporting requirements, safety protocol disclosure mandates, and oversight structures across major commercial jurisdictions. That's real patchwork. It's the strongest FOR-side data point we've seen in months.

Here's the counterargument we have to take seriously, because we think we may be underweighting it. The NDAA and intelligence authorization vehicles have historically been how tech governance achieves bipartisan passage when the direct path is blocked by partisanship. CFIUS reform (FIRRMA, 2018) is the cleanest example: years of standalone legislation going nowhere, then passage embedded in the FY2019 NDAA with overwhelming bipartisan support once the national security frame dominated. Export control updates followed the same pattern. The AI OVERWATCH Act has passed the House Foreign Affairs Committee — that's not a press release, it's an actual procedural milestone. If national security framing consolidates behind a must-pass vehicle the way FIRRMA did, our 18% is probably 10 points too low. We're not revising it today because the current evidence shows committee passage of one bill, not the floor coalition dynamics that characterized FIRRMA's final push. But this is the part of our thesis that genuinely keeps us up at night: we may be discounting the must-pass vehicle path because we're anchoring too heavily to the failed standalone legislation history.

What would move us? Above 30%: the AI OVERWATCH Act's provisions are formally incorporated into the NDAA markup with bipartisan floor support before the August recess — that would signal the FIRRMA playbook is actually executing. Below 10%: the September legislative calendar closes without a vehicle and the midterm positioning phase begins in earnest, at which point passage before November becomes structurally impossible regardless of political will. One thing this 18% explicitly does not account for: whether the Q2 earnings cycle produces enough public AI-attribution language from major employers to shift congressional urgency. If companies start publicly naming AI as a displacement driver — which our white-collar displacement forecast at 73% suggests is increasingly likely — that adds constituent pressure that could change the Senate calculus faster than any White House framework document.