A bill introduced in the House on August 6, 2026 would tax the use of foundation models and tie the rate directly to the national unemployment rate. H.R. 10044, the AI Tax and Work Protection Act, sets a base excise tax of 2% on token value that escalates automatically as unemployment climbs, doubling its rate of increase once unemployment passes 7%. The revenue would fund job-creation grants through a new Work Protection Administration inside the Department of Labor.
Key facts
- H.R. 10044, introduced August 6, 2026 by Representative Greg Casar, with Representatives Valerie Foushee and Sara Jacobs as cosponsors.
- The tax is the greater of token value times an applicable token percentage or transaction value times an applicable transaction percentage.
- The token rate starts at 2% and rises one point per point of unemployment above 5%, then two points per point above 7%; the transaction rate starts at 3% on the same schedule.
- Primary source: the full bill text on GovInfo.
Most legislative responses to AI and employment have been studies, commissions, and reporting requirements. This one is a tax with a formula, and the formula is the interesting part. The rate is not fixed by Congress and it is not set by an agency. It is a function of a number the Bureau of Labor Statistics publishes every month.
Work through what that does. At 5% unemployment or below, the token rate is 2%. Between 5% and 7%, it becomes 2% plus the excess, so 6% unemployment means a 3% rate. Above 7%, the multiplier doubles: 9% unemployment produces a rate of 2% plus twice the two-point excess, or 6%, three times the base. The transaction-based alternative runs the same schedule from a 3% floor, and taxpayers pay whichever of the two produces the larger figure.
The design intent is not subtle. If AI deployment displaces workers at scale, the thing doing the displacing gets progressively more expensive, and the proceeds go to a fund for putting people back to work. It is an automatic stabilizer aimed at a specific technology, closer in structure to a carbon price that ratchets with emissions than to an ordinary sales tax. The bill establishes a trust fund and, in its own language, a "Work Protection Administration" within the Department of Labor to administer job-creation grants from it.
The mechanism has a real elegance and a real problem, and they are the same feature. Tying a rate to unemployment means Congress does not have to predict how fast AI displaces labour, which is fortunate, because nobody can. But it also means the tax responds to unemployment from any cause. A recession driven by interest rates, a supply shock, or a pandemic would raise the AI tax rate just as reliably as a wave of automation would, and it would raise the cost of the technology precisely when businesses are least able to absorb new costs. The bill treats unemployment as a proxy for AI-driven displacement, and it is a proxy that has been wrong about the cause of joblessness for most of American economic history.
There is a second implementation question the token base raises directly. Taxing on token value assumes tokens are a stable, measurable unit of AI consumption, which was roughly true in 2023 and is getting less true every quarter. Models that spend more tokens to think harder would be taxed more heavily than models that produce the same answer tersely, which is a strange incentive to write into tax law. The alternative transaction base exists in the bill presumably as a hedge against exactly this, and taking the greater of the two suggests the drafters expected each to be evadable in different ways.
Why this matters even though it will almost certainly not pass: introduced bills with three sponsors are how policy positions get drafted into concrete language, and concrete language is what later bills copy. The AI-and-labour debate has been conducted almost entirely in the abstract, in op-eds and hearings about whether displacement is real. H.R. 10044 is one of the first attempts to write down a specific number, a specific base, a specific escalation schedule, and a specific agency. Whatever happens to this bill, that text now exists and can be argued with in detail rather than in principle.
The honest caveat is the size of the gap between this and law. A House bill with a handful of cosponsors, a new excise tax, a new federal administration, and an industry with substantial lobbying resources on the other side is not close to enactment. Read it as a marker of where part of the Democratic caucus is heading, not as a forecast of your future API bill.
Originally published on Ground Truth, where every claim is checked against the primary source.
Top comments (0)