The IRS processed roughly 163 million individual tax returns in fiscal year 2024. By the end of 2026, the number of enforcement employees responsible for examining, collecting, and investigating those returns is on pace to drop below 30,000, a level that would fall short of any point during President Trump’s first term, when the IRS was already considered dangerously understaffed.
That projection is based on workforce reduction data published by the Treasury Inspector General for Tax Administration (TIGTA) through March 2025, layered onto the staffing baseline in the IRS’s own fiscal year 2024 Data Book. No single agency report stamps a precise future headcount, but the trajectory is stark: separations are outrunning hiring, and the gap is widening.
At the same time, the agency has quietly formalized rules governing artificial intelligence systems that help decide which taxpayers get audited. The collision of those two trends raises a practical question for millions of filers: Who reviews the work when the machines flag your return and the humans are gone?
How the workforce got here
TIGTA’s snapshot report catalogs the mechanisms driving rapid headcount declines: probationary-period terminations, a deferred resignation program, and attrition that accelerated sharply starting in late 2024. The report does not forecast a single future number, but the separation rates it documents are severe enough that, applied to the FY2024 Data Book baseline of enforcement division staffing, they point toward a sub-30,000 workforce within months.
For context, IRS Data Books from Trump’s first term (fiscal years 2017 through 2020) show examination, collection, and criminal investigation personnel hovering in the low-to-mid 30,000s. Those levels drew bipartisan criticism. In 2022, Congress responded by approving roughly $80 billion in new IRS funding through the Inflation Reduction Act, with a large share earmarked for hiring enforcement agents and modernizing technology.
That hiring surge never fully materialized. Subsequent budget deals clawed back billions in IRA funding, and the positions that were filled have now collided with the current wave of departures. The net result is a workforce trajectory that could leave the agency with fewer enforcement staff than it had during the very period that prompted the emergency funding in the first place.
AI moves from theory to agency policy
While the human workforce contracts, the IRS has been building out the rulebook for its AI-assisted enforcement tools. Internal Revenue Manual section 10.24.1 establishes a formal governance framework for artificial intelligence and machine learning systems used in tax administration. The policy explicitly covers AI tools that influence audit selection and scope, requiring ethical reviews, bias assessments, and oversight board approval before high-impact models go live.
The manual traces its authority partly to Executive Order 13859, a 2019 directive on maintaining U.S. leadership in AI, and to provisions in the fiscal year 2023 National Defense Authorization Act that required federal agencies to inventory and govern their AI systems.
What the policy does not reveal is equally important: how many AI models are currently in production, how many audits they influence, or what their error rates look like. IRM 10.24.1 is a governance framework, not a performance report. But its existence confirms that AI-driven audit selection is operational enough to require its own chapter of the tax code’s internal playbook. The IRS is not experimenting with machine learning in a lab. It is writing rules for systems already embedded in the process that decides whether your return gets a second look.
What this means if you get flagged
For taxpayers who file straightforward returns and never hear from the IRS beyond a refund deposit, the staffing decline may be invisible. The trouble starts when something goes wrong.
Consider a self-employed graphic designer who claims a home-office deduction and reports income from a dozen 1099 clients. An AI model scores the return as high-risk based on patterns in the data, perhaps an unusual ratio of expenses to income or a mismatch with third-party filings. The system routes the case to an examination queue. From there, a human agent is supposed to open the file, send a notice requesting receipts and contracts, review whatever the filer sends back, and issue a determination. Before the staffing decline, that cycle might have taken a few months. With fewer agents handling a potentially larger pool of machine-selected cases, the same filer could wait considerably longer for each step, all while interest accrues on any disputed amount.
The National Taxpayer Advocate, in her 2025 Annual Report to Congress, put the risk in practical terms. The Advocate found that taxpayer service metrics were strong during the 2025 filing season, with improved call wait times and faster return processing. But the report warns that those gains may not survive into 2026 as experienced employees leave and remaining staff absorb increasingly complex workloads. Payment plan negotiations, identity-theft cases, and disputed audit findings all require seasoned personnel, and those are exactly the roles being hollowed out.
Where the public record stands as of May 2026
Four primary government sources, each covering a different piece of the picture, form the foundation of what the public currently knows. The TIGTA snapshot report documents the pace and mechanisms of workforce loss. The FY2024 Data Book provides the most recent official baseline for staffing and enforcement activity. IRM 10.24.1 confirms that AI is a governed, operational component of audit selection. And the Taxpayer Advocate’s 2025 report translates those trends into a practical forecast for filers.
Individually, none of these documents confirms that enforcement staffing has already crossed below 30,000, or that AI now drives a specific percentage of audit selections. The sub-30,000 projection comes from combining TIGTA’s documented separation rates with the Data Book baseline, a reasonable and transparent inference, but not a figure any single agency has published. The expansion of AI in enforcement is confirmed by the existence of detailed governance rules, not by published metrics on case volume or accuracy.
The IRS has not released an updated headcount or a strategic plan explaining how it intends to balance growing automation against accelerating staff losses. Until it does, the situation amounts to this: the agency is losing experienced people at a documented pace, it is formalizing the use of AI tools that help decide who gets audited, and its own internal watchdogs expect the combination to create real problems for any filer who needs a human being on the other end of the line.
