Does the IRS use AI to select tax returns for audit?

Yes. GAO report GAO-26-107522 (published March 24, 2026) found that the IRS runs 126 active AI use cases, up from 54 in 2024. These include AI-powered anomaly detection for audit selection, pattern matching against historical return data, and document verification systems. The GAO report also flagged concerns about 'unintentional algorithmic biases' in IRS AI systems — a finding that has compliance implications for both the IRS and for tax preparers advising clients on audit risk.

What are the known AI audit triggers for IRS audit selection in 2026?

Based on IRS technical documentation and tax professional reporting in 2026, the AI audit selection system is known to flag: (1) significant year-over-year income discrepancies without supporting documentation, (2) deduction ratios that fall outside industry norms for the reported SIC code, (3) excessive use of round numbers (which correlates with estimation rather than documentation), (4) underreported self-employment income flagged by document cross-referencing, and (5) inconsistencies between 1099s and W-2s and the reported amounts on the return. Importantly, the AI system also conducts pattern matching against networks of related taxpayers — an unusual deduction strategy that correlates with known audit cases will flag returns it appears in.

Should accounting firms change how they prepare returns because of IRS AI?

Yes, with specificity. Three adjustments are warranted: (1) Documentation now explains anomalies proactively — if a client's income dropped 40% year-over-year for a legitimate reason, a brief explanation in supporting documentation is cheap insurance against AI flagging. (2) Precision beats estimation — avoid round numbers and approximate deductions where documentation supports specificity. (3) Reconcile 1099s and W-2s before filing — AI document matching is faster and more comprehensive than prior manual processes. These are adjustments at the margin, not wholesale changes to return preparation methodology.

What does the GAO report say about algorithmic bias in IRS AI systems?

GAO-26-107522 flagged that some IRS AI systems may exhibit 'unintentional algorithmic biases' — patterns in audit selection that correlate with taxpayer characteristics beyond income reporting behavior. The GAO recommended IRS implement bias testing across its AI use cases. For tax professionals, the practical implication is limited: the bias findings apply to IRS governance of its systems, not to changes in return preparation. However, if clients from demographic groups historically subject to higher audit rates ask about the IRS AI findings, this is the accurate framing: the GAO identified a concern and recommended IRS testing, not a confirmed systematic bias in a specific audit outcome.

IRS AI Audit Selection: CPA Firm Guide 2026

Summary

GAO report GAO-26-107522 (March 24, 2026) confirmed IRS runs 126 active AI use cases, up from 54 in 2024, including AI-based audit selection
Known AI audit triggers: income discrepancies, deduction ratio outliers, round numbers, underreported self-employment income, document mismatches
The IRS also runs 4.8 million taxpayer interactions through AI voice bots — the scope of AI in IRS operations is broader than most CPA firms realize
Three return preparation adjustments for accounting firms: document anomalies proactively, use precise figures not estimates, reconcile all 1099s and W-2s before filing

What the GAO Found

On March 24, 2026, the Government Accountability Office published report GAO-26-107522: a comprehensive review of artificial intelligence use at the Internal Revenue Service.

The headline finding: the IRS operates 126 active AI use cases as of the report date — more than double the 54 cases identified in a 2024 review. The expansion covers audit selection, taxpayer service (AI voice bots handling 4.8 million taxpayer calls), fraud detection, collections prioritization, and document processing.

The finding that matters most for accounting firms: the IRS's audit selection system uses AI anomaly detection to identify returns that deviate from statistical norms — and that system has been significantly expanded.

The GAO also flagged a concern that received less coverage: the report identified potential "unintentional algorithmic biases" in IRS AI systems and recommended that the IRS implement formal bias testing protocols across its AI use cases. The IRS agreed to the recommendation.

How IRS AI Audit Selection Works

The IRS has not published its full AI audit selection methodology — and the specifics of their models are proprietary. What is known from IRS technical documentation, GAO findings, and reporting from tax practitioners is that the system operates on several well-documented principles:

Statistical Outlier Detection

AI models compare each return against a statistical distribution of returns from similar taxpayers — same income range, same entity type, same reported industry (SIC code). Returns where specific line items fall outside normal ranges for that comparison group receive elevated scores in the audit selection process.

This is not new — the DIF (Discriminant Income Function) score has operated on similar principles for decades. What AI adds is faster processing across more variables simultaneously and the ability to incorporate external data (1099s, W-2s, real estate records, bank data where available) in real time rather than through periodic matching.

Pattern Matching Against Known Cases

AI systems flag returns that share characteristics with previously audited cases — including characteristics beyond obvious financial patterns. A deduction strategy that appeared in a cluster of cases that resulted in adjustments will flag other returns that include similar patterns, even if those returns are from unrelated taxpayers.

The practical implication: aggressive deduction positions that have been successfully audited against in the past will continue to generate audit flags even when properly documented.

Document Mismatch Detection

The IRS receives copies of 1099s, W-2s, K-1s, and other third-party information returns from employers and financial institutions. AI systems match these documents against the amounts reported on taxpayer returns. Discrepancies — even minor ones that result from legitimate rounding or timing differences — can trigger flags.

The AI processes these matches faster and more comprehensively than prior manual systems. Returns that might previously have slipped through a manual matching process are more likely to be flagged.

Three Return Preparation Adjustments for Accounting Firms

These adjustments do not require changing professional judgment about tax positions. They are documentation and presentation practices that reduce the probability of AI-flagged anomalies that are, in fact, legitimate.

Adjustment 1: Document Anomalies Proactively

If a client's return has a legitimate anomaly — income dropped significantly because of a business setback, a large deduction resulted from a qualifying event, a year was unusual for a documented reason — include a brief explanation in supporting documentation.

The AI system flags anomalies relative to the prior year and relative to industry norms. A documented explanation does not guarantee the flag won't be raised, but it changes the risk profile: the explanation is part of the record, the return preparer has identified the issue, and the audit inquiry has an immediate, professional answer.

The habit to build: For every return with a significant year-over-year change in income, deductions, or credits, add a one-paragraph documentation memo explaining the change.

Adjustment 2: Precision Beats Estimation

Round numbers correlate with estimation rather than documentation. Deductions of $5,000, $10,000, and $25,000 are statistically associated with estimated rather than documented amounts — and AI models are trained to recognize this pattern.

Where documentation exists, use the precise documented amount. Where expenses are partially estimated (home office, vehicle use, some mixed-use categories), document the calculation methodology explicitly rather than rounding to a convenient number.

This is a documentation discipline, not a change in legal position.

Adjustment 3: Reconcile All 1099s and W-2s Before Filing

Before filing any return, confirm that every third-party information document received by the client is accounted for in the return — either as reported income or with a documented explanation for why the amount differs (incorrect 1099, nominee distribution, etc.).

The AI document matching system processes these reconciliations faster than prior manual processes. A 1099 mismatch that might have taken two years to surface through a manual compliance check may now generate an inquiry within weeks of filing.

What This Means for Your Advisory Practice

The IRS AI expansion creates a service opportunity for accounting firms that move first.

An AI audit risk review — an advisory engagement that reviews a client's return for AI-flagged anomalies before filing — is a new service category that clients with complex returns will pay for. The value proposition is clear: a review by a CPA who understands AI audit selection before the IRS AI reviews the return, at a cost significantly lower than the cost of an audit.

For CPAs who want to build this service, the three adjustments above are the starting framework. The premium version incorporates industry-specific deduction benchmarks and year-over-year variance analysis against the client's historical returns.

Sources

GAO Report GAO-26-107522: Artificial Intelligence at the Internal Revenue Service, March 24, 2026
IRS Strategic Plan: AI Use Cases and Data Governance, 2025–2026
American Institute of CPAs: Tax Section Alerts, March 2026

The Crossing Report covers the AI transition for professional services firm owners. The premium tier includes a CPA firm AI audit risk review checklist, benchmarked against 2026 IRS audit selection patterns. Subscribe here →

IRS AI Audit Selection: What the GAO Report Means for Accounting Firms (2026)