Two new reports: Fair Tax Collection Practices and managing AI risks Inbox Treasury Inspector General for Tax Administration Unsubscribe 8:18 AM (3 hours ago) to me TIGTA Fiscal Year 2026 Statutory Review of Potential Fair Tax Collection Practices Violations Why did we do this audit? The IRs Restructuring and Reform Act of 1998, Section 7803 (d)(1)(G), requires us to report semiannually to Congress on administrative or civil actions involving Fair Tax Collection Practices (FTCP) violations, such as contacting taxpayers at an unusual time or place, harassment or abuse, or bypassing an authorized representative. We reviewed IRS employee violations of the FTCP provisions in the Internal Revenue Code § 6304 and reported or potential Fair Debt Collection Practices Act violations by private collection agency employees, including any related administrative or civil actions. What did we find? We identified 32 potential FTCP violations in the IRS’s Automated Labor and Employee Relations Tracking System (ALERTS) between July 1, 2024, and June 30, 2025. Most of the violations involved contacting a represented taxpayer and bypassing a duly authorized taxpayer representative. In 16 of these violations, the case was correctly coded as an FTCP violation. We identified the remaining 16 FTCP violations through additional research in ALERTS that found they were incorrectly coded as other types of misconduct. In our Fiscal Year 2025 FTCP review, we previously recommended that the IRS enhance training to ensure that miscoding cases involving potential FTCP violations are correctly coded and resolved according to IRS policies and procedures. Read the report Assessment of Artificial Intelligence Risk Management Practices Why did we do this audit? Office of Management and Budget (OMB) Memorandum M 25-21 requires federal agencies to implement minimum risk management practices for high-impact artificial intelligence (AI) use cases by April 2026. Because AI systems can pose data quality and reliability risks, agencies should prioritize oversight based on the assessed risk level and potential impact. We evaluated the IRS’s process to document its data reliability assessments for its AI use cases. What did we find? The IRS has used AI for many years, including for tax return classification and issue selection. As of December 2025, the IRS had 225 AI use cases. We reviewed five presumed high-impact use cases and found that the IRS did not fully document: Evaluation of risk. Two of the five use cases (40 percent) lacked documented impact assessments. The remaining 60 percent (3 of 5) of AI use cases did have a documented impact assessment. As of July 2026, Treasury had not issued guidance on minimum risk management practices. The IRS developed and is implementing its own AI governance policy, and processes to align with OMB Memorandum M-25-21. Quality and appropriateness of the data. Although data quality checks were performed, documentation varied. Twenty percent (1 of 5) of the cases had documents outlining numerous processes and procedures used in testing. Eighty percent (4 of 5) use cases lacked testing documentation, and the IRS had not standardized documentation or established baseline procedures for evaluating data quality before using data in high-impact AI models.

Two new reports: Fair Tax Collection Practices and managing AI risks Inbox Treasury Inspector General for Tax Administration Unsubscribe 8:18 AM (3 hours ago) to me TIGTA Fiscal Year 2026 Statutory Review of Potential Fair Tax Collection Practices Violations Why did we do this audit? The IRs Restructuring and Reform Act of 1998, Section 7803 (d)(1)(G), requires us to report semiannually to Congress on administrative or civil actions involving Fair Tax Collection Practices (FTCP) violations, such as contacting taxpayers at an unusual time or place, harassment or abuse, or bypassing an authorized representative. We reviewed IRS employee violations of the FTCP provisions in the Internal Revenue Code § 6304 and reported or potential Fair Debt Collection Practices Act violations by private collection agency employees, including any related administrative or civil actions. What did we find? We identified 32 potential FTCP violations in the IRS’s Automated Labor and Employee Relations Tracking System (ALERTS) between July 1, 2024, and June 30, 2025. Most of the violations involved contacting a represented taxpayer and bypassing a duly authorized taxpayer representative. In 16 of these violations, the case was correctly coded as an FTCP violation. We identified the remaining 16 FTCP violations through additional research in ALERTS that found they were incorrectly coded as other types of misconduct. In our Fiscal Year 2025 FTCP review, we previously recommended that the IRS enhance training to ensure that miscoding cases involving potential FTCP violations are correctly coded and resolved according to IRS policies and procedures. Read the report Assessment of Artificial Intelligence Risk Management Practices Why did we do this audit? Office of Management and Budget (OMB) Memorandum M 25-21 requires federal agencies to implement minimum risk management practices for high-impact artificial intelligence (AI) use cases by April 2026. Because AI systems can pose data quality and reliability risks, agencies should prioritize oversight based on the assessed risk level and potential impact. We evaluated the IRS’s process to document its data reliability assessments for its AI use cases. What did we find? The IRS has used AI for many years, including for tax return classification and issue selection. As of December 2025, the IRS had 225 AI use cases. We reviewed five presumed high-impact use cases and found that the IRS did not fully document: Evaluation of risk. Two of the five use cases (40 percent) lacked documented impact assessments. The remaining 60 percent (3 of 5) of AI use cases did have a documented impact assessment. As of July 2026, Treasury had not issued guidance on minimum risk management practices. The IRS developed and is implementing its own AI governance policy, and processes to align with OMB Memorandum M-25-21. Quality and appropriateness of the data. Although data quality checks were performed, documentation varied. Twenty percent (1 of 5) of the cases had documents outlining numerous processes and procedures used in testing. Eighty percent (4 of 5) use cases lacked testing documentation, and the IRS had not standardized documentation or established baseline procedures for evaluating data quality before using data in high-impact AI models.