The AI Tax and Work Protection Act is a bill in the 119th Congress that would impose a tax on artificial-intelligence token usage and establish a Work Protection Administration within the Department of Labor. It was introduced in the House of Representatives on August 6, 2026 by Representative Greg Casar (D-Texas), chair of the Congressional Progressive Caucus, with Representatives Valerie Foushee and Sara Jacobs. Its stated purpose is "to impose a tax on artificial intelligence token usage and establish a Work Protection Administration within the Department of Labor, and for other purposes" (H.R. 10044 — AI Tax and Work Protection Act; Source: casar.house.gov). The bill was referred to the Committee on Education and Workforce and additionally to the Committee on Ways and Means. A section-by-section account of the introduced text is at H.R. 10044 — AI Tax and Work Protection Act.
The tax
The instrument is an excise tax, added as a new section 4491 in a new subchapter D of chapter 36 of the Internal Revenue Code rather than as a general corporate levy (H.R. 10044 — AI Tax and Work Protection Act). It attaches to a "covered person" — one who develops a foundation model, sells access to one, or modifies an existing open-weight foundation model — and who either earns revenue from a covered transaction or uses the model to reduce their own workforce. A covered transaction is the provision of model access to an unrelated party in a trade or business, or use of a model by the taxpayer or a related party where that use "enables or results in a reduction in the workforce."
The tax is assessed on two bases and takes the larger: the bill "measures both tokens sold and revenue on AI sales" and taxes "the higher of those two." At the base tier the rates are 2 percent of the fair market value of tokens processed in covered transactions, and 3 percent of consideration received for AI services plus the fair market value of covered transactions with related parties (H.R. 10044 — AI Tax and Work Protection Act). Casar has described the two-base structure as an anti-avoidance design, intended to stop companies "gaming the system by rapidly changing the definition of a token" — a token being a unit set by the seller rather than by statute or by any external standard. The bill answers that by directing the Secretary to determine the fair market value of a token in consultation with the Secretary of Commerce. See Inference Economics and Token Pricing on token pricing as the industry's billing metric.
The rate is contingent on labor-market outcomes rather than fixed. In Casar's account the tax "starts out quite low, when we don't have A.I. mass unemployment, and then ramps up if A.I. starts to cause significant job loss" (Source: nytimes.com). The text keys that ramp to the Bureau of Labor Statistics U-4 measure rather than to headline unemployment, using the highest U-4 of the three calendar years preceding the transaction. Above 5 percent both rates rise point for point with the excess; above 7 percent they rise at twice the excess. The Secretary may, with the Secretary of Labor, discount unemployment attributable to "a war, pandemic, or any other massive economic shock unrelated to the use of artificial intelligence" (H.R. 10044 — AI Tax and Work Protection Act). The triggering logic — an obligation keyed to a measured displacement threshold rather than to a date — matches the tiered proposals surveyed at AI Labor Disruption, applied here to the revenue side.
Coverage turns on a statutory compute threshold. A "foundation model" is defined as one trained on broad data, generally by self-supervision, applicable across a wide range of contexts, and trained using at least 10^25 integer or floating-point operations — or such quantity as the Secretary determines necessary for comparable capability in a given calendar year (H.R. 10044 — AI Tax and Work Protection Act). See Compute Thresholds. Research and development use by federal, state and local governments, institutions of higher education, Federally Funded Research and Development Centers and 501(c)(3) organisations is excluded. The tax applies to transactions occurring more than one year after enactment.
The Work Protection Administration
Revenue would fund a Work Protection Administration inside the Department of Labor, which would award job-creation grants to states, cities, tribes, localities, nonprofits, unions and educational institutions. The named categories of work are building housing, modernizing infrastructure, and improving access to child care and elder care. All receipts flow to a Treasury trust fund from fiscal year 2027, usable only for the administration and the grant program, with up to 20 percent authorised each year for Workforce Innovation and Opportunity Act programs and training (H.R. 10044 — AI Tax and Work Protection Act).
The text enumerates seventeen grant categories in section 402(b), broader than the three the sponsor's account highlights. Beyond housing, infrastructure and care they include public education, health initiatives, homelessness services, community violence prevention, scientific research support, public art and archives, environmental conservation and climate resilience, wildfire prevention and disaster preparedness, school and community-facility repair, parks and public lands, and local news and journalism — the last covering local news funds, journalism fellowships and public-interest news gathering capacity (H.R. 10044 — AI Tax and Work Protection Act).
Grantees must meet labour conditions: policies preserving section 7 rights under the National Labor Relations Act, recognition of an exclusive representative where more than 50 percent of employees select one, a local hiring preference, prevailing-wage pay, health coverage comparable to the Federal Employees Health Benefits Program, 12 workweeks of paid family and medical leave, and accrued paid leave. Funded hiring may not lay off or partially displace existing employees or replace workers on strike. Tribal governments are exempt from the eligibility policy requirements (H.R. 10044 — AI Tax and Work Protection Act).
The bill also directs the Bureau of Labor Statistics to report on AI's workforce impacts beyond displacement, expressly including the degradation of existing jobs through lower pay, fewer hours, or conversion to temporary work, and authorises $20 million a year for that purpose from FY2027 through FY2031 (H.R. 10044 — AI Tax and Work Protection Act). An advisory committee drawn from Labor, Commerce, Education, the Federal Reserve, the NLRB, NSF, the Council of Economic Advisers and OSTP, together with union representatives, state governors and two labour economists, would meet quarterly to advise on the program.
The text is internally inconsistent about the name of the body it creates: Title III is headed "Office of Job Creation at Department of Labor," section 301 establishes the "Work Protection Administration," and section 407(2) defines the Director as "the Director of the Office of Job Creation" (H.R. 10044 — AI Tax and Work Protection Act).
Casar has defended routing the revenue to job creation rather than to a small universal basic income, calling a small UBI "dystopic" — the objection being to a low-value cash transfer as a substitute for employment rather than to income support as such (Source: nytimes.com). See Universal Basic Income (in AI policy).
Status
Introduced August 6, 2026 and referred the same day to the Committee on Education and Workforce and to the Committee on Ways and Means. One version of the text exists, Introduced in House. No further action is established by the available record (H.R. 10044 — AI Tax and Work Protection Act).
Relationships
- related: AI Labor Disruption — the displacement premise the tax rate is keyed to.
- related: Universal Basic Income (in AI policy) — the alternative use of revenue the sponsor rejects.
- related: Inference Economics and Token Pricing — tokens as the metered unit the tax would attach to.
- depends-on: H.R. 10044 — AI Tax and Work Protection Act — the text as introduced.
- related: Compute Thresholds — the 10^25 operations threshold defining a taxable foundation model.
- related: Greg Casar — sponsor.