The IRS Is Using AI to Pick Audit Targets. Oklahoma Business Owners Should Be Paying Attention.

I need you to update your mental model of how IRS audits work. The old paradigm — a human agent reviews a stack of returns, flags anomalies by hand, sends you a letter — is still technically true, but it's increasingly a downstream step in a process that starts with algorithms you'll never see.

As of April 2025, the IRS reported 101 active projects involving artificial intelligence. That number has only grown. These projects span customer service, operations, and — the part that should concern you — enforcement. Specifically, the IRS has integrated statistical and machine-learning techniques into the process used to select tax returns for audit. The goal is to reduce "no-change" audits (audits that find nothing wrong), which means the returns the IRS does select are increasingly the ones where the algorithm has already identified a probable discrepancy.

Translation: if the IRS comes knocking in 2026, they probably already know what they're looking for.

What the AI Is Actually Doing

The IRS's automated systems cross-reference your return against an expanding universe of third-party data: W-2s, 1099s, payroll records, bank information reports, and starting in 2026, Form 1099-DA for digital asset activity. The machine-learning layer compares your reported figures against industry norms, historical patterns, and peer benchmarks. If your Oklahoma restaurant reports a 12% profit margin when the national average for restaurants is 6-9%, the system notices. If your S-Corp distributions spike while your reported salary stays flat, the system notices. If your gross receipts don't match your 1099-K filings within a certain tolerance, the system definitely notices.

This isn't speculation. TIGTA (the Treasury Inspector General for Tax Administration) has been documenting these capabilities in annual reports. The IRS's own Strategic Operating Plan explicitly targets partnership audits at 1% (up from 0.1% in 2019), high-income individual audits at 16.5% for those earning over $10 million, and large corporate audits at 22.6% for entities with assets over $250 million.

The Oklahoma Angle

Here's what my Oklahoma clients need to understand: the IRS doesn't audit geography. It audits patterns. But certain patterns are disproportionately common in Oklahoma's economy. Oil and gas royalty owners with complex mineral interests. Cannabis operators (yes, they exist, and yes, the IRS is watching even though federal legality is murky). S-Corp shareholders who've been a little too aggressive with their salary-to-distribution ratios. Multi-entity real estate investors using 1031 exchanges without proper qualified intermediary documentation.

If you fit any of those profiles, the AI isn't going to give you the benefit of the doubt. It's going to flag the pattern and hand it to a human reviewer who will decide whether to open an examination.

The Paradox: Fewer Agents, More Targeted Audits

The IRS lost approximately 25% of its workforce last year through voluntary separations and retirements. A federal hiring freeze complicated recruitment. Multiple acting commissioners cycled through leadership. By any measure, the agency is understaffed. But here's the paradox that most taxpayers misunderstand: fewer agents doesn't mean fewer audits. It means more targeted audits. When you have less human capacity, you lean harder on automation to identify the highest-value targets. The returns that do get selected are increasingly the ones with the biggest expected yield.

For my clients in the $400K-and-above income range, the audit rate is going up, not down. The IRS has explicitly stated that taxpayers below $400K won't see increased scrutiny relative to historical levels. That's the commitment. Above $400K, all bets are off — and the machine-learning models are getting better every cycle.

What to Do About It

The best defense is a return that doesn't trigger the algorithm in the first place. That means: accurate third-party matching (your 1099s should reconcile perfectly), defensible deduction ratios (if you're claiming home office, vehicle, and meals that exceed 20% of gross receipts, you'd better have bulletproof documentation), and — if you're an S-Corp owner — a reasonable compensation analysis that you can hand to an auditor on day one.

I prepare audit defense packages for all of my S-Corp clients as part of annual planning. Not because I expect an audit. Because if the algorithm flags you, I want the response ready before the letter arrives.

If you haven't had someone stress-test your return against current IRS patterns, schedule a consultation. This isn't about fear. It's about being prepared.