AI tax and work protections

Full Title:
AI Tax and Work Protection Act

Summary#

This bill would create a new excise tax on certain uses of large artificial intelligence models (called “foundation models”) and use the money to fund a new Work Protection Administration (WPA) inside the Department of Labor and a national jobs grant program. The tax rate would rise when unemployment is higher. The WPA would run competitive grants to hire people for public-interest jobs (child care, education, health, infrastructure, conservation, etc.) with worker protections and wage/benefit rules.

  • Main change: adds a new excise tax on “covered persons” that develop, sell access to, or modify foundation models and that earn revenue from or use those models in ways that reduce workforce levels.
  • New agency and program: creates a Work Protection Administration to award grants to eligible public and nonprofit entities to create full‑time jobs in specified public services.
  • Worker protections: jobs funded by the grants must meet minimum wage/prevailing wage rules, offer health coverage comparable to federal employees, provide paid family leave and paid time off, and follow nondisplacement rules.
  • Trust fund and use of revenue: tax receipts go to a trust fund that must be used to run the WPA and the jobs program; up to 20% of trust funds can be used for Workforce Innovation and Opportunity Act training.
  • Data and oversight: the Bureau of Labor Statistics is directed to collect data on AI’s effects on workers and is authorized $20 million per year for several years to do so.

What it means for you#

  • AI companies and developers

    • If you develop, sell access to, or modify a foundation model and generate revenue from those activities or use them to reduce staff, the company would owe an annual tax based on tokens processed or transaction value.
    • The tax rate starts at a small percentage (2% for token measure; 3% for transaction measure) and steps up if unemployment rises above 5%.
    • You would face new reporting, valuation, and compliance requirements; the IRS must issue rules on how to value tokens.

  • Workers

    • New jobs created with grant funds must be permanent, full-time where possible, pay at least applicable minimum or prevailing wage, offer health benefits, paid family leave, and paid time off accrual.
    • Grants include nondisplacement rules intended to prevent replacing existing workers with grant-funded hires.

  • State and local governments, schools, nonprofits, tribal governments, and institutions of higher education

    • These entities are eligible to apply for competitive grants to hire staff in many public-service areas (child care, education support, public health, elder care, housing, infrastructure, conservation, disaster preparedness, local journalism, research support, and more).
    • Priority is given to projects creating permanent, full‑time jobs and to state/local governments when the work is typically a government function.
    • Tribal governments are eligible but are exempt from some of the grant program’s employment policy requirements.

  • Taxpayers and the federal budget

    • The bill directs tax revenue from the new AI excise to a trust fund to pay for the WPA and grants. The bill itself does not contain an overall dollar estimate of revenue or total program cost.

  • Researchers and the public

    • The Bureau of Labor Statistics will collect and report data on AI’s broader impacts on work, beyond simple job loss, with dedicated funding for several years.

Expenses#

Estimated public cost: No overall revenue or budget estimate is provided in the bill text.

  • The bill creates a trust fund to receive the excise tax revenue and says those funds shall be used to run the WPA and the grants program. The bill does not state total annual spending amounts for the WPA or grants.
  • The Bureau of Labor Statistics is authorized $20 million per year for each fiscal year 2027 through 2031 for collecting data on AI impacts.
  • Up to 20% of funds appropriated to the trust fund may be used for Workforce Innovation and Opportunity Act programs and training.
  • Administrative costs to create and run the Work Protection Administration and to implement the tax (IRS rulemaking, enforcement, valuation of tokens) are not estimated in the bill text.
  • No fiscal note or concrete revenue projections are included in the provided material.

If you need precise cost or revenue numbers, the bill text does not provide them. No publicly available information.

Proponents' View#

A possible argument for the bill, based on the bill text, is:

  • The bill appears intended to make companies that profit from large AI models pay into a fund that supports workers and public services when AI reduces employment.
  • Tying the tax rate to the unemployment rate could target higher charges when labor markets are weaker and job loss risks may be greater.
  • Using revenue to create public-service jobs aims to replace lost private-sector work with community-focused employment (child care, health, infrastructure, etc.).
  • Requiring strong wages, benefits, paid leave, and nondisplacement protections seeks to ensure that grant-funded jobs are good, stable jobs.
  • Investing in BLS data collection is meant to improve knowledge about how AI affects jobs beyond displacement (hours, pay, job quality).

Opponents' View / Concerns#

Based on the bill’s design, reasonable concerns include:

  • One concern is that the bill’s method for identifying and measuring taxable activity is complex. Determining the “fair market value” of a token, counting tokens processed, and proving that use “enables or results in” workforce reduction could be hard to implement and audit.
  • The definition of “foundation model” contains a numerical compute threshold written as “10\25\” in the text provided. It is unclear what precise compute level that represents, and the bill says the Secretary may set comparable thresholds for later years. This makes the scope of models covered unclear.
  • The tax could raise compliance and administrative costs for companies and for the IRS. Those costs are not estimated in the bill text.
  • The tax might be translated into higher prices for customers, reduced investment in AI development, or other business responses; the bill text does not analyze these potential effects.
  • It is unclear how causation would be proven for uses that “enable or result in a reduction in the workforce,” which could lead to disputes or avoidance strategies (for example, shifting work classifications or transaction structures).
  • The advisory committee is exempted from the Federal Advisory Committee Act, which may reduce public transparency about its deliberations.
  • The bill does not include an overall estimate of how much revenue will be raised or the total cost of the grant program and new administration. This leaves uncertain whether tax receipts will cover program ambitions or require additional appropriations.

What is unclear: the exact numerical compute threshold for “foundation model,” how tokens will be valued in practice, how IRS and Labor will measure workforce reductions tied to AI use, and the likely scale of tax revenue and program spending. The bill text does not provide those details.