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

Case study

An IRS watchdog finds missing AI risk documentation, with a dated remediation commitment

What happened

TIGTA—the Treasury Inspector General for Tax Administration—published report 2026-208-059 ↗ on September 24. It examined documentation for five deployed high-impact AI uses supporting IRS workload selection, drawn from seven identified relevant uses. Two lacked documented impact assessments and four lacked performance-testing documentation. Managers said testing had occurred. The audit did not evaluate the accuracy of the underlying data itself or measure erroneous tax decisions.

IRS management accepted recommendations on impact assessments and standard data-quality documentation. It committed to completing assessments for deployed high-impact uses as of November 2026, and said the standard data-quality process was already implemented. That management statement is not a verified remediation outcome. This is a documentation/control audit, not a representative performance study of the IRS's entire AI inventory.

Why it matters

An established public watchdog can expose missing safeguards inside a revenue agency. Documentary oversight improves visibility, but finding missing records is different from proving that a model is inaccurate or that the agency has corrected the problem.

Supporting sources

  1. tigta.gov · Source 1 ↗