AI Governance Institute
← News
Enforcement2026-09-29

IRS Deployed High-Impact AI With No Testing Records in 80% of Cases

Source

Data documentation for IRS's AI use cases is lacking, watchdog finds

Treasury Inspector General for Tax Administration

What happened

TIGTA published a report, covered by FedScoop in "Data documentation for IRS's AI use cases is lacking, watchdog finds". It found that 80 percent of five sampled high-impact IRS AI use cases lacked any written testing documentation. The IRS was conducting data quality checks in practice. However, those checks were not being recorded in the AI impact assessments that OMB Memorandum M-26-04: Increasing Public Trust in AI Through Unbiased AI Principles requires agencies to complete. TIGTA flagged that undocumented testing creates conditions where biased, inaccurate, or unreliable AI outputs could go undetected and uncorrected. IRS management agreed with both TIGTA recommendations: standardize data quality assessment procedures across AI use cases, and finish impact assessments for all currently deployed high-impact systems by November 2026.

Why it matters

  • ·Federal agencies and contractors subject to OMB AI inventory and assessment requirements now have a watchdog-confirmed standard: performing checks is not enough if those checks are not recorded. Undocumented testing is treated the same as no testing under OMB Memorandum M-26-04: Increasing Public Trust in AI Through Unbiased AI Principles.
  • ·The finding exposes a documentation gap that regulators and auditors are beginning to treat as a material control failure, not a paperwork technicality. Organizations that cannot produce written evidence of their AI data quality and testing procedures face the same reputational and regulatory exposure the IRS now does.
  • ·The November 2026 remediation deadline creates a concrete benchmark. Private-sector teams building AI impact assessment programs can use it to pressure-test whether their own documentation would survive an equivalent review.

Stay ahead of stories like this

Get every US AI governance development like this one, plus the rest of the week's developments. Every Thursday.

Free every Thursday. Unsubscribe anytime.

Governance controls affected

ALC-002High-Risk AI Audit TrailALC-005Regulatory Audit ReadinessMON-001AI Performance BaselineMGV-006RAI Benchmark-Aligned Evaluation FrameworkHOC-001AI System Risk Classification

What to do now

  • ☐Pull your current AI use case inventory and confirm that each high-impact use case has a written impact assessment on file, not just an internal understanding that testing occurred.
  • ☐Ask the teams running AI data quality or validation checks to show you the written records from those checks. If the records do not exist, begin creating a documentation requirement immediately.
  • ☐Set a review date before November 2026 to verify that any AI systems without completed impact assessments have been brought into compliance, mirroring the IRS remediation deadline.
  • ☐Check whether your organization's AI documentation standards require sign-off from a named owner for each use case, so that informal practices cannot persist without a responsible party on record.
  • ☐If your organization holds federal contracts or is subject to OMB AI guidance, confirm with your legal or compliance team whether the TIGTA findings trigger any parallel obligations for your own AI use case inventory.

What to watch next

Compliance teams should watch for TIGTA follow-up reviews as the November 2026 IRS deadline approaches. A missed remediation will likely draw a second public finding. It could also influence how other inspectors general treat AI documentation gaps across the federal government. Broader OMB enforcement signaling around OMB Memorandum M-26-04: Increasing Public Trust in AI Through Unbiased AI Principles is also worth tracking. Of particular interest is whether the administration treats documentation failures as grounds for pausing AI deployments. The TIGTA finding joins a pattern visible in the NY Comptroller audit of SUNY. Public-sector watchdogs are consistently finding that AI deployment has outpaced the governance infrastructure meant to oversee it.

Related Coverage

Research2026-11-08

Boards Flag AI Ownership Gap as Grant Thornton Survey Finds Controls Undocumented

Grant Thornton's survey of executives and board members found that AI governance and AI use rank among the top sources of anticipated regulatory scrutiny. Board participants specifically identified unclear ownership and decision rights as central problems. The firm urged organizations to document governance processes, maintain AI audit trails, and preserve evidence that controls are actually operating.

Research2026-11-04

Anthropic's FlashProve Lets Enterprises Prove Which AI Model Answered

Anthropic has published FlashProve, an open-source prototype that uses cryptographic proofs to verify that an AI model's output came from a specific declared model. The system processes over 100 tokens per second on a single GPU for a 400-billion-parameter model and is released under an open-source license. It directly addresses the enterprise audit gap created when AI responses are routed through third-party intermediaries or when model versions change without disclosure.

Research2026-11-03

1,000+ AI Rules, One Audit Gap: Runtime Evidence Is What Regulators Want

Openlayer's Country-by-Country AI Regulation Guide maps more than 1,000 AI policy initiatives across 72 countries. It finds that regulators worldwide are converging on the same core requirements: risk classification, human oversight, transparency, post-market monitoring, and version-linked documentation. The critical finding for compliance teams is that audit readiness now requires evidence that controls operated during live production use, not just that they were designed.