In 2024, the IRS was running 54 artificial intelligence and machine learning models to analyze tax returns and flag enforcement targets. By 2026, that number is 125.

The IRS used to find audit targets by sampling returns and relying on human reviewers. Now they run a targeting system that processes every return against industry benchmarks, third-party data, prior years, and cross-referenced information sources — in seconds. The same dollar of unpaid tax is no harder to owe. It’s substantially easier to find.

Here’s what these models specifically flag.

Income That Doesn’t Match What Was Filed About You

The most common AI trigger, by a wide margin, is a discrepancy between what you reported and what a third party reported about you. When a client sends a 1099-NEC, a platform issues a 1099-K, or an employer files a W-2 — that data hits the IRS before your return does. The AI runs a match. If your return shows less income than the sum of forms filed in your name, that mismatch is flagged immediately.

The “it’s just a few thousand dollars” logic doesn’t hold. The AI doesn’t skip small gaps. And now there’s a new one to watch: Form 1099-DA, the first-ever broker reporting form for digital assets, required for 2025 transactions forward. If you sold, exchanged, or disposed of cryptocurrency and your broker filed a 1099-DA, that number is in the IRS system. Report less than they have, or nothing, and you have an automatic discrepancy.

If you have cryptocurrency transactions from 2025 or later — even a single trade — assume the IRS already has the paperwork on it.

Cash-Heavy Businesses with Gaps Between Deposits and Revenue

Restaurants, contractors, auto repair shops, nail salons, and any business with a high volume of cash get elevated scrutiny — not because owners are assumed to be cheating, but because cash income is harder to cross-reference than electronic payments. The AI compensates by looking harder at what can be verified.

The specific technique: a bank deposit analysis. Your total bank deposits are compared to your reported gross receipts. If you deposited $420,000 and reported $290,000 in revenue, the gap gets flagged. The IRS isn’t automatically assuming fraud — but they’re going to ask how you explain $130,000 in deposits that don’t appear in your income. Personal transfers, loans, and capital contributions can explain some of it, but you need documentation for every dollar of the difference.

Running a cash business in New York and keeping sloppy records is a higher-risk position in 2026 than it was two years ago. Clean separation between your business account and personal account is your first line of defense. See also: what commingling business and personal funds looks like to an auditor.

If your cash-heavy business has inconsistent recordkeeping, the time to clean it up is before a notice arrives — not after. Talk to us about what you should be tracking.

S-Corp Owners with a Salary That Looks Wrong Against Their Distributions

The IRS has targeted low S-Corp officer salaries for years. The AI makes it faster to catch at scale. The model compares your W-2 officer wages to your industry, your own prior-year compensation, and your total distributions. If you’re pulling $200,000 out of the business annually and paying yourself $12,000 in salary, the AI marks that pattern for review.

The consequence of getting this wrong is expensive. The IRS can reclassify excess distributions as wages retroactively, then assess back payroll taxes — plus penalties and interest from the original due dates. In a significant reclassification, that bill can run $30,000–$60,000 or more by the time penalties compound.

The standard: your salary should reflect what you’d pay a third party to do the same work in your business. Most owner-operators in professional services land in the $60,000–$120,000+ range before distributions. If you’re below that threshold while taking substantial distributions, you’re in a risk zone the AI is built to surface.

More detail: S-Corp reasonable salary — how to pay yourself without inviting an audit.

Deductions That Sit Outside Your Industry’s Normal Range

The AI benchmarks your deductions against comparable businesses in your industry with similar gross revenue. If your vehicle deductions are three times what a typical business in your category takes, or your meals expense is running at 20% of revenue when the industry average is 4%, those outliers get flagged.

This doesn’t mean you should underclaim deductions that are legitimate. It means your documentation needs to actually exist and actually support the numbers. A vehicle used exclusively for business with clean mileage logs is fine even if it exceeds the average. A vehicle deduction without a mileage log or business-purpose record is not a documentation problem — it’s an exposure.

Schedule C filers are at statistically higher audit risk than corporations to begin with. The AI layer increases that gap. See why Schedule C returns get audited more than S-Corps for the data.

Worker Misclassification

The IRS trains its models on contractor-versus-employee patterns specifically because misclassification is one of the largest gaps in payroll tax collection. If you’re paying someone on a 1099-NEC who works exclusively for you, follows your schedule, uses your equipment, and is directed by you on how to do their job — that’s an employee by the IRS’s behavioral control test, not a contractor.

The penalty structure for misclassification is steep: unpaid payroll taxes on every dollar paid to the misclassified worker, plus penalties that typically add 15–40% on top of that. In a multi-year audit covering three or four workers reclassified as employees, the total can run well into six figures.

If you have ongoing workers you’re paying on 1099, it’s worth a quick review of whether they’d pass the behavioral test. The worker misclassification rules cover this in detail.

What You Can Actually Do

The AI doesn’t evaluate intent. It evaluates data patterns. Your protection is clean, documented, explainable numbers — not smaller numbers.

Report everything that was reported about you. Separate your business and personal accounts. Pay yourself a defensible salary from your S-Corp. Document deductions at the time they happen. If you have cryptocurrency, report it.

If your records are inconsistent or your deduction patterns are unusual, now is the time to address that — before a notice starts the clock on a response deadline.

Not sure whether your return has exposure to one of these patterns? Schedule a review before the IRS finds it first. We work with clients across Long Island and nationwide.

This post reflects publicly available IRS enforcement priorities and AI deployment data as of June 2026. The IRS does not publish its specific audit selection algorithms. The information here is based on IRS published data, congressional testimony, and publicly known enforcement trends. This is general educational information, not legal or accounting advice. If you have received an IRS notice, consult a qualified tax professional immediately. Geiger Tax & Accounting — Amityville, NY — (631) 532-5622 — geigertax.taxdome.com.