AI Regulation Tracker / Agency enforcement practice

How the IRS Uses AI to Pick Partnership Audits

The IRS already runs two AI models to prioritize which large partnerships get audited, the broader enforcement push targets high-income filers and complex partnerships, and in 2026 it is expanding AI across enforcement even as its staff shrinks. Here is what CPAs and high-net-worth advisors should understand about an audit pipeline that is becoming algorithmic.

By Anthony Guerriero · Reviewed by The Leveraged Years Editorial Desk · Published June 28, 2026 · Last updated June 28, 2026

How the IRS Uses AI to Pick Partnership Audits

The Leveraged Years AI Regulation Tracker

The IRS already uses artificial intelligence to help decide who gets audited. The Government Accountability Office reported in June 2024 that the IRS "currently uses two AI models to help prioritize partnership returns for audit," intended to surface the highest-risk large partnership returns, and in 2026 the agency told Congress it is expanding AI across enforcement even as its staff shrinks. The model logic is not published, so for CPAs and high-net-worth advisors the practical exposure is being selected without a stated reason. Primary source: U.S. GAO, "Artificial Intelligence May Help IRS Close the Tax Gap" (June 6, 2024).

Regulatory Briefing

Bindingness: Agency enforcement practice. This is operational use of AI by the IRS to select returns for audit, not a new statute or regulation. It changes how the agency works, not the tax law you owe.
Scope: United States federal tax enforcement / Primarily large and complex partnerships and high-income, high-wealth filers, with GAO also reporting AI use in the IRS's annual research audits and in selecting some refundable-credit claims for review.
What Changed: The IRS has moved audit selection toward machine learning. The Government Accountability Office confirmed the IRS "currently uses two AI models to help prioritize partnership returns for audit," intended to surface the highest-risk large partnership returns. The agency built this out under its 2023 enforcement push aimed at high earners and partnerships, and in 2026 it told Congress it is expanding AI across enforcement even as staffing falls.
Who This Affects: Large partnerships and the CPAs, tax directors, and high-net-worth advisors who prepare and defend their returns, plus any high-income filer in the IRS focus segments. The model logic is not published, so the practical exposure is being selected without a clear, stated reason.
Current Status: In use and expanding
Effective Date: AI models in use for partnership audit selection as of the GAO's 2023 and 2024 reporting, with the IRS describing continued expansion in its fiscal year 2027 Congressional Justification (April 2026)
Primary Source: U.S. GAO, "Artificial Intelligence May Help IRS Close the Tax Gap" (June 6, 2024): https://www.gao.gov/blog/artificial-intelligence-may-help-irs-close-tax-gap
Workflow Impact: Audit selection is shifting from human judgment toward a risk score you cannot see. The defensible response is documentation built before a return is selected: contemporaneous substantiation, clean and reconciled balance sheets, and explanations for the items a model is most likely to flag, so that a return that gets pulled survives the exam that follows.
Operational Next Step: For every large partnership and high-income client, treat the file as if it will be scored. Reconcile beginning and ending balance sheet figures, attach the required statements that explain differences, document the basis for large or unusual positions, and keep substantiation organized at filing time rather than assembling it after a notice arrives.
Last Verified: June 28, 2026

The agency that audits you is now using AI to decide who gets audited

The most consequential AI rollout for high-income filers this year is not a new product. It is inside the IRS. The agency that decides whether your return gets examined is using machine learning to help make that decision, and it is doing so first in the corner of the tax world where the dollars are largest and the returns are hardest to read: large partnerships.

The Government Accountability Office, the federal watchdog that reviews how agencies work, put it plainly. In a June 2024 report on the tax gap, GAO wrote that the IRS "currently uses two AI models to help prioritize partnership returns for audit," and that "the models are intended to help select the highest risk large partnership returns for audit." That is not a pilot or a press release. It is the federal auditor's own description of how returns are being chosen right now.

For a CPA or a wealth advisor, that reframes the work. The question is no longer only whether a position is defensible if examined. It is whether the return looks like the kind of return a model scores as high risk in the first place.

Why partnerships, and why now

The IRS did not pick partnerships at random. GAO noted that the number of large partnerships grew by nearly 600 percent between 2002 and 2019, that these returns are complex, and that when the IRS does audit them the result is often no change. That combination, high volume, high complexity, and low yield from traditional selection, is exactly the problem machine learning gets pointed at.

The enforcement push behind it is on the record. In September 2023, in release IR-2023-166, the IRS announced what it called a sweeping effort to shift attention toward high-income earners, partnerships, and large corporations. It described expanding its Large Partnership Compliance program "with the help of AI," using what it called "machine learning technology" to identify compliance risk, and it opened examinations of 75 of the largest partnerships in the country, each with more than 10 billion dollars in assets on average. The same release committed the IRS to not raising audit rates on filers earning less than 400,000 dollars a year. The design intent is concentration at the top.

What the models actually do, and what no one outside the IRS can see

It is worth being precise about what is known, because the temptation is to overstate it. The public record says the IRS uses two AI models to prioritize partnership returns and aims them at the highest-risk large returns. It does not publish the features those models weigh, the thresholds they use, or the exact factors that raise a score. Anyone claiming to know the inputs is guessing.

What the IRS has signaled in plain terms is one risk category it is watching: balance sheet discrepancies. In IR-2023-166 the agency flagged partnerships over 10 million dollars in assets that show differences between end-of-year balances and the next year's beginning balances, often without the required statement explaining the difference, and called that an indicator of potential noncompliance. That is not the model's full logic, but it is a documented signal the agency itself named, and it is the kind of clean, checkable discrepancy that a data-driven selection system can flag automatically.

The transparency problem is the IRS's, and it is also yours

GAO did not simply describe the models. It flagged a weakness. The watchdog wrote that the IRS "could do more to document these AI models" so they are transparent and applied consistently, and warned that "otherwise, IRS risks not being able to explain how it selected taxpayers for these more burdensome audits." GAO made formal recommendations to improve the documentation and transparency of the partnership models, so the IRS can show how it selected the returns it pulls.

Read that from the taxpayer's side. If the agency cannot fully explain why a model selected a return, the filer who gets pulled cannot be told why either. The due-process and transparency limits here are real, and they are not yet resolved. That is not a reason to panic. It is a reason to make every return able to stand on its own, because the selection logic is not something you can argue with in advance.

2026: more AI, fewer people

The trend is not slowing. In March 2026, GAO published a follow-up report finding the IRS had 126 active AI use cases as of the prior summer, spanning taxpayer services, operations, and compliance and fraud detection, including audit selection. IRS officials told GAO they plan to keep expanding AI use. In the agency's fiscal year 2027 Congressional Justification, summarized in April 2026 reporting, IRS leadership said the agency is "modernizing enforcement through expanded use of artificial intelligence, advanced analytics, and improved data integration" to "more precisely identify high-risk noncompliance."

That expansion is happening against a shrinking workforce. GAO reported that the roughly 79.4 billion dollars the IRS received in 2022 to modernize has been cut to about 25.9 billion, and that the agency lost roughly 20 percent of its staff in 2025, including 63 employees in the group that supports AI development and oversight. The Taxpayer Advocate Service counted the workforce falling from more than 102,000 at the start of 2025 to about 74,000 by December. Fewer examiners and more automated selection point the same direction: the front door of an audit is increasingly an algorithm, and the human review comes after you are already inside.

What CPAs and advisors should do before a return is scored

The work is not new in kind. It is the same substantiation and reconciliation a good preparer already does, moved earlier and made deliberate, because the file is now effectively scored at filing.

Reconcile the balance sheet. For partnerships over the thresholds the IRS named, make sure beginning and ending balances tie out, and attach the required statement when they do not. This is the one signal the agency has publicly identified.
Substantiate the large and the unusual. The items that sit outside the norm for a return's size and industry are the ones most likely to draw a score. Document the basis for them at filing, not after a letter arrives.
Keep the file exam-ready. Organize support so a selected return can withstand the examination that follows. Selection you cannot prevent. A clean defense you can build.
Do not try to game an unseen model. The inputs are not public, and reverse-engineering a score you cannot see is a waste of effort. Filing a complete, consistent, well-supported return is the strategy that works whether or not a model pulls it.

The shift here is quiet but real. For the segments the IRS is focused on, audit selection is becoming algorithmic, and the people who advise large partnerships and high-net-worth clients are the ones who will field the questions when a return gets pulled. The professionals who treat every such return as if it will be scored are the ones who will be ready when it is.

Source File

Primary Source: U.S. Government Accountability Office, "Artificial Intelligence May Help IRS Close the Tax Gap" (June 6, 2024), reporting that the IRS currently uses two AI models to prioritize partnership returns for audit: https://www.gao.gov/blog/artificial-intelligence-may-help-irs-close-tax-gap
Primary Source: IRS, "IRS announces sweeping effort to restore fairness to tax system" (IR-2023-166, Sept. 8, 2023), on the AI-leveraged Large Partnership Compliance expansion and high-income focus: https://www.irs.gov/newsroom/irs-announces-sweeping-effort-to-restore-fairness-to-tax-system-with-inflation-reduction-act-funding-new-compliance-efforts
Primary Source: U.S. Government Accountability Office, "Inside the IRS's Use of Artificial Intelligence" (March 25, 2026), on 126 active AI use cases, planned expansion, and funding and staffing cuts (report GAO-26-107522): https://www.gao.gov/blog/inside-irs-use-artificial-intelligence
Last Checked: June 28, 2026
Status: In use and expanding
Editorial Note: This is workflow-level analysis for tax and finance professionals, not legal, tax, or financial advice. Confirm how IRS enforcement practice applies to a specific client with qualified counsel.

Frequently Asked Questions

Is the IRS really using AI to select audits?

Yes. The Government Accountability Office reported in June 2024 that the IRS "currently uses two AI models to help prioritize partnership returns for audit," intended to select the highest-risk large partnership returns. GAO reports that AI also helps select returns in other compliance programs, including the IRS's annual research audits and some refundable-credit claims, such as the Earned Income Tax Credit.

Does the IRS publish how its audit AI works or what raises a risk score?

No. The IRS has not published the features its partnership models weigh or the exact factors that raise a score. GAO flagged this, warning the IRS "risks not being able to explain how it selected taxpayers for these more burdensome audits," and recommended the agency better document the models. The one risk signal the IRS has named publicly is unexplained balance sheet discrepancies on large partnership returns.

Who is most affected by AI audit selection?

Large and complex partnerships and high-income, high-wealth filers, the segments the IRS targeted in its 2023 enforcement push. The agency committed to not raising audit rates on filers earning under 400,000 dollars a year. The practical exposure for affected filers and their advisors is being selected by a model whose logic is not disclosed.

What should a CPA or advisor do about it?

Treat every large-partnership and high-income return as if it will be scored at filing. Reconcile balance sheets and attach required explanatory statements, document the basis for large or unusual positions, and keep substantiation organized so a selected return survives the exam. The inputs are not public, so the goal is a complete and defensible return, not gaming an unseen model.