Accounting

The IRS’s AI Era: What It Means for Your Next Return

By Mark G. McCreary, partner, chief artificial intelligence & information security officer at Fox Rothschild LLP On Jan 12, 2026

Artificial intelligence has moved from pilot to practice at the IRS. The agency is using machine learning and data analytics to spot risk faster, sift through vast records, and guide exam resources. For businesses and their advisors, this shift will change who gets audited, how exams unfold, and what preparation looks like before filing. It is not science fiction. It is operations.

The Big Shift: From Legacy Scores to Pattern Recognition

For years, the IRS leaned on legacy scoring models that weighed known red flags. Those rules worked, but they were blunt. They also struggled to keep pace with new structures, pass-through complexity, and cross-entity flows. AI changes the lens. Models now look for patterns across returns, forms, payments, and even third-party data. That means more signal, less noise. More accuracy is likely to follow.

Expect a sharper focus on high-risk cases, especially those with layered entities, unusual loss positions, or mismatched information returns. At the same time, there should be fewer no-change audits. Better selection increases the chance that an audit results in an adjustment. There is a strong argument that this is good policy. It is also a cue to businesses. If you are selected, the IRS likely sees something specific. You should too.

The audit experience will shift as well. Agents will have tools that surface risk factors and anomalies in real time. Examinations may narrow faster. The agency can route complex cases to specialized teams sooner. Delays tied to manual screening should fall. Documentation quality will matter more than ever.

What to Do Now, Before You File

Start with alignment. Reconcile your story across the return, financials, and footnotes. Inconsistencies are easy for machines to spot. They are harder to explain after selection. Tie key tax positions to contemporaneous memos. Cite the specific code sections and facts relied upon. Keep those memos in a ready file.

Clean your data. Many risk signals arise from simple errors. Mismatched names or EINs. Round-dollar entries that do not reconcile. Missing attachments. Incomplete partner or shareholder information. Run a prefile check that mirrors what an algorithm might do. Match 1099s and W-2s. Validate intercompany charges. Confirm that book-to-tax bridges and schedules tie.

Document transfer pricing, valuation, and R&D claims with care. These areas attract model attention because the amounts are large and the facts are subtle. Record supply chains, method selection, and comparables. Keep working papers that link to the filed return and to invoices or contracts.

Anticipate questions. If you have a new structure, a large NOL, or a change in method, draft a short explanation statement. Clear context lowers the risk that the return looks like an outlier. It also speeds any review.

Coordinate with your advisors early. Agree on the audit narrative before filing. Map who owns which facts. Store support in a single index, with version control. If you face an exam, the first 30 days matter. Good organization reduces scope creep.

Is AI a Fix for Staffing, or a New Operating Model

Both, but mainly the latter. The IRS has lost experienced staff in recent years. AI helps triage workload and route talent where it matters most. Yet the larger change is cultural. The agency is moving from static rules to adaptive detection. It is also building feedback loops. As cases close, models learn which signals predicted adjustments and which did not. That cycle gets better with time. Once started, it does not depend on headcount alone.

Emerging Issues You Should Watch

Transparency is tightening. The IRS will show outcomes, but it will not expose the guts of its models. That is standard for risk systems. Still, it means fewer clues about thresholds. Businesses should not rely on historical audit triggers.

Human oversight remains central. Agents make determinations. But tools will shape focus. Ask early for clarity on scope and rationale. Put your facts on the record. If you think a model flag rests on a data error, say so and provide proof.

Bias and fairness need testing. AI can reflect patterns in its training data. If that data over-represents a type of entity or industry, selection can skew. Expect the IRS to keep refining controls. You should track whether your sector sees unusual attention and raise concerns through industry groups when warranted.

Privacy-by-design is advancing. More analytics means more data handling. Safeguards, role-based access, and audit trails are essential. The agency is under pressure to show strong controls. That is good for taxpayers. It also raises the bar on your own data hygiene. Keep TINs, payroll records, and K-1 details secure.

A Brief Word on Personal Impact

Business owners and executives file individual returns tied to complex businesses. AI will link those filings more effectively. Flow-through mismatches, basis errors, and large charitable or real estate items may draw more review. Coordinate your personal and business advisors. Align the facts on both returns.

In Conclusion

AI at the IRS is here to stay. It will reduce no-change audits, sharpen the focus on high-risk items, and speed the exam process. Success now depends on clean data, coherent narratives, and rigorous documentation. Treat your return like it will be read by a machine first and a human next. If you do, you will lower your audit risk. If you are selected, you will be ready.

About the Author:

Mark G. McCreary is partner, chief artificial intelligence & information security officer at Fox Rothschild LLP.

To access more business news, visit NJB News Now.