On March 24, 2026, the Government Accountability Office published GAO-26-107522, a review of how the Internal Revenue Service manages artificial intelligence.
The IRS reported 126 active AI use cases as of June 2025, up from 10 in August 2022. Sixty-five were too sensitive for public reporting or were exempt research projects. Seventy-seven were still in development. Most aimed at operational efficiency or at tax compliance and fraud detection.
The inventory had gaps. More than a quarter of the use cases did not say how they would benefit the agency. The inventory also left out some AI-enabled tools, including several that IRS officials said were contracted to help build criminal cases.
No one managed the whole. Several offices oversee individual use cases, but none manages AI investments across the agency, and the IRS has no process to make sure those investments support its goals.
The workforce has thinned. GAO said 2025 staffing reductions could greatly affect the agency’s ability to use AI. The research and analytics unit lost 63 employees who had worked on AI. The IRS has not identified the AI skills it needs or a plan to close the gaps, even as it plans to expand AI use.
GAO made eight recommendations, including a skills-gap plan, quality checks on the inventory, inventory coverage of contracted and law enforcement AI, coordination across business units, and outcome metrics for each use case. The IRS agreed with all eight. As of October 2026, GAO listed all of them as open, with an IRS status letter expected this fall.
The IRS can count its AI projects. GAO found it cannot yet say what all of them are for, who manages them together, or who will run them.
Why it matters
The findings are specific to one agency. The recommendations work as a checklist for any organization with a growing AI inventory. An inventory that leaves out contracted tools misses AI the organization depends on. A use case with no stated benefit cannot be measured. And a portfolio with no owner grows by accumulation, not decision.
The workforce finding is the hardest to fix. AI projects need people who understand them. When those people leave, a project can lose the expertise that built it.
Questions for your organization
- Does your AI inventory include AI inside contracted tools and services?
- Does every use case state the benefit it should deliver, and is that benefit measured?
- Who manages AI investments across the whole organization?
- Have you identified the AI skills you need, and who has them today?
- When a key employee leaves, which AI projects lose their owner?
An inventory counts projects. Governance decides which ones should exist.
Sources