The Race — Building Democracy in the Age of AI. People race toward a shared finish line.
The Race
← Specific Events

United States — federal tax administration

Watchlist

The AI Tax Integrity Act would require a Treasury pilot and an independent check of its accuracy

What happened

Today's recovery resolves much of the old **D26-15** legislative lead. Representative Vern Buchanan introduced **H.R. 9501 on June 29**; Ways and Means approved an amended version **July 1, 40–0**, and the [reported text](https://www.govinfo.gov/content/pkg/BILLS-119hr9501rh/html/BILLS-119hr9501rh.htm) was printed **September 8**. It would require Treasury to begin an AI pilot within 180 days of enactment, operating for 18 months to two years. The pilot would identify inaccurate returns involving identity theft, fraudulent claims or improperly identified preparers. Within 180 days after it ends, the Government Accountability Office would report detected fraud, actual recoveries and tool accuracy to congressional tax committees. The amended operative text goes beyond the bill's older study-oriented title. It does not establish a new IRS enforcement authority. A later official floor-action or enactment record was not recovered; September 8 is the latest authoritative action verified here.

The [Congressional Budget Office estimate](https://www.cbo.gov/publication/62715), dated **August 20** and reposted September 10, estimates less than $500,000 in appropriated implementation costs over 2026–31, no direct-spending change and a revenue effect below $500,000 whose direction cannot be determined. CBO expects much of the work is already covered by executive actions. This is not evidence of a large fiscal windfall or a measured accuracy improvement.

For The Race, the distinct mechanism is an operational tax-administration test paired with an outside audit of results. It could build public implementation knowledge, but procurement dependence, error rates and taxpayers' challenge rights will determine its power implications. This is fiscal capacity through administration, rather than a tax on AI-generated surplus. It also supplies a concrete proposed example for the existing **Independent testing and evaluation** tool, accurately labelled as proposed.

Potential direction of power

Undetermined

Effects depend on retained public knowledge, vendor dependence, accuracy and taxpayer challenge; proposal alone does not establish direction.

Window of contestability

Unclear

Unclear until binding pilot/oversight design and actual implementation are verified.

Supporting sources

  1. govinfo.gov · Supporting source 1 ↗
  2. govinfo.gov · Supporting source 2 ↗
  3. cbo.gov · Supporting source 3 ↗