Senator Bernie Sanders introduced legislation proposing direct payments to Americans funded by artificial intelligence revenue, while President Trump issued an executive order mandating federal review of foundation models. The twin moves mark the first serious policy architecture around AI wealth capture since the technology crossed into commercial deployment at scale.
Sanders' proposal targets the estimated $1.2 trillion to $3.4 trillion in annual economic value AI systems are projected to generate by 2030, per Treasury Department modeling. The legislation would establish a federal trust funded by a levy on AI-generated corporate revenue above baseline thresholds, with quarterly distributions to citizens meeting income caps. Trump's order, issued without prior comment from the Office of Management and Budget, directs the National Institute of Standards and Technology to catalogue all foundation models with parameter counts above 10 billion and assess national security implications within 90 days. The review includes mandatory disclosure of training data sources, a provision that puts every frontier lab's intellectual property framework under federal microscope.
The Sanders mechanism is narrow but durable. It does not tax compute or model weights. It taxes the revenue those models generate once deployed in commercial settings, which means enforcement rides on existing IRS infrastructure rather than new regulatory apparatus. That makes it harder to kill in reconciliation and easier to expand once the precedent exists. The income cap ensures initial benefits flow to households earning below $120,000, a threshold that covers 72% of U.S. tax filers. The trust structure isolates the revenue stream from annual appropriations, which is how Alaska's oil dividend survived four decades of shifting political majorities.
Trump's model review carries different second-order effects. Mandatory disclosure of training data sources forces every lab to document whether copyrighted material, classified information, or foreign datasets entered the training corpus. That creates legal surface area for IP litigation and export control enforcement that did not previously exist. The 90-day window means preliminary findings land in late June, ahead of any legislative markup on AI governance bills currently stalled in Senate Commerce. The order also establishes parameter count as a regulatory threshold, which has never been formalized in U.S. policy. Models above 10 billion parameters now sit in a different compliance category, and every lab with a roadmap toward 100 billion or 1 trillion parameter models must now assume tiered federal oversight.
Allocators should track three specific developments. First, whether Sanders attaches the AI revenue levy to must-pass reconciliation vehicles expected in Q3, which would force a Congressional Budget Office score on the revenue potential and make the proposal's fiscal math public. Second, whether NIST's model catalogue includes foreign-deployed models accessible to U.S. users, which would extend compliance obligations to non-U.S. entities and trigger jurisdictional disputes with the EU's AI Act enforcement regime. Third, whether Treasury's September 15 fintech risk assessment, already mandated under separate executive action, incorporates AI-generated trading strategies as a systemic risk category, which would layer capital requirements onto any fund using algorithmic models above the parameter threshold.
The Sanders proposal has 11 Senate co-sponsors, none of whom sit on Finance Committee, which means it currently lacks markup pathway. The Trump order has no sunset provision and survives any staff turnover at NIST.