IRS AI Audit Enforcement: What It Means for Your CPA Practice — and the Advisory Opportunity Hidden Inside It

As of the June 2025 inventory — the most recent publicly confirmed count — the IRS runs 126 active AI use cases, up from 54 in 2024. That growth rate tells you something: the IRS is deploying AI aggressively across tax administration, not just piloting it quietly.

The use cases that matter most for CPA firm owners:

Audit case selection. AI systems analyze return data to flag anomalies — patterns that, in historical audit data, have been associated with underreporting or non-compliance. The model does not read your client's cover note explaining why their Schedule C looks unusual this year. It reads the pattern and flags it.

Known flag triggers from public IRS guidance and GAO analysis:

• Year-over-year income changes without documentary support
• Deduction ratios that are statistically extreme for a given income level and industry
• Round-number entries across schedules where precise actual figures would be expected
• Mismatches between third-party reporting (1099s, K-1s, W-2s) and what the return shows
• Self-employment income that doesn't reconcile against known third-party data sources

Third-party document reconciliation. IRS AI automatically compares documents received from third parties — 1099s, K-1s, broker statements — against the filed return. A mismatch doesn't wait for a human. It triggers an automated flag.

Taxpayer correspondence. More than 4.8 million taxpayer calls are handled by IRS AI voice bots. If your client calls about a notice, there is a meaningful probability they are talking to an AI before — or instead of — a human.

What GAO-26-107522 also found: the IRS has been expanding AI use rapidly without commensurate oversight. The GAO flagged that "unintentional algorithmic biases" may be affecting audit selection in ways the IRS hasn't identified or corrected. That phrase matters: some taxpayer categories may be getting flagged at higher rates because of patterns in the training data, not because of actual compliance risk.

---

The Documentation Gap: Why Most Client Returns Are Underexplaining Themselves

Here's the core problem most CPA firms haven't named yet.

The IRS AI doesn't read intent. It reads patterns. When a return has a large deduction, an unusual income swing, or a complex entity structure, the AI sees a flag-worthy pattern — regardless of whether the underlying transaction is completely legitimate and well-documented. The difference between a client who gets audited and a client who gets the audit resolved quickly is what was in the file before the notice arrived.

Most CPA firms prepare accurate returns. Far fewer prepare returns that are preemptively documented against what an AI system is trained to flag. That gap is the advisory opportunity.

What AI-flaggable patterns look like — and what explains them:

Large deduction in a low-deduction-ratio category. A sole proprietor home-based consulting firm with $180,000 income and $47,000 in business equipment expenses. The deduction ratio is unusual. What explains it: Section 179 election in a specific year, documented equipment list, business purpose notation. Without the attachment, the AI flags the ratio. With it, the file explains itself before review.

Significant year-over-year income swing. A real estate LLC with K-1 distributions that swung from $85,000 to $210,000 in a single year. The AI flags the change. What explains it: a documented property sale or refinancing event with backup documentation in the workpapers. The K-1 itself doesn't carry that context. The file has to.

Owner salary that looks anomalous for an S-corp structure. An S-corp owner paying themselves $52,000 in salary against $400,000 in total distributions. The IRS has long targeted unreasonably low S-corp salaries. The AI version of this scrutiny is faster. What explains it: documented compensation study or industry comparables with a brief notation in the return workpapers showing the salary was set deliberately.

Precision matters at the line-item level, too. AI audit models have been trained on data where round numbers correlate with estimation, and estimation correlates with underreporting. A Schedule C with $4,000 in office supplies looks different to the model than one showing $3,847. Not because of the dollar difference — because precision signals documentation. Where actual figures exist, use them.

---

The New Advisory Service You're Not Offering (But Should Be)

Call it what it is: AI audit-proofing.

The service is a pre-filing documentation depth review. Before a return with potentially anomalous characteristics is filed, a qualified reviewer goes through the workpapers and confirms that every AI-flaggable pattern has an explanatory attachment. It is positioned to clients as: "We review your return the way the IRS AI will — and we make sure the file explains itself before it gets flagged."

This is not a new kind of accounting. It's a new framing of what good CPA firms have always done. What's changed is that you can now name it, price it, and market it against a specific and urgent external threat your clients are aware of.

What the service includes:

1. Pattern identification: Review the return for any element that statistically matches IRS AI flag triggers — income changes, deduction ratios, round numbers, reconciliation gaps, entity structure anomalies.
2. Documentation completeness check: For each flagged element, confirm the workpapers include an explanatory attachment — a note, a supporting document, a business purpose statement, or a reconciliation schedule.
3. Gap remediation: For any flag trigger without adequate documentation, work with the client to produce it before filing. Not after a notice.
4. Filing confirmation: A brief written sign-off in the file that the AI audit-proofing review was completed.

Pricing model:

• Flat-fee add-on: $200–$500 per complex return, depending on firm size and return complexity. Position as an optional but strongly recommended add-on for any return with anomalous characteristics.
• Bundled advisory retainer: For business clients on monthly or quarterly advisory retainers, include AI audit-proofing as part of the annual filing service. Raise the retainer 10–15% to cover it.

The marketing line that works: "We review your return the way the IRS AI will — and we make sure the file can answer back."

This is concrete. It's specific. And it gives clients a reason to pay for something they previously assumed was included in the base service.

---

The Enforcement Timeline: What to Tell Clients Now

The March 2026 GAO report contains a nuance that changes the client conversation — but only if you handle it correctly.

Short-term (2026): IRS AI enforcement is weaker than projected. The 63 AI-skilled employees lost from the RAAS division are the people who built, validated, and maintained those 126 AI use cases. Without them, the GAO found the risk that IRS AI programs "will not succeed" is materially elevated. The AI-powered audit selection model — designed to identify complex underreporting patterns in high-income and multi-entity returns — may be significantly degraded or delayed.

Medium-term (2027–2028): The IRS will rebuild. The underlying data infrastructure is intact. The agency still has budget authority for AI initiatives. Rebuilding human AI expertise in a government context typically takes 18–36 months. The most realistic scenario: IRS AI audit capability resumes at higher effectiveness in 2027–2028 than it had in 2025.

The posture for your clients: The near-term urgency has decreased. The risk posture has not. The right advice is not "the IRS isn't looking." It's "their most sophisticated looking capability got delayed — which gives you time to get your documentation right before it resumes."

How to frame this in client conversations:

"The IRS lost the team that was building their AI audit systems this year. Their capability is weaker right now than where it was heading. That's not a reason to take more risk — it's a reason to use the window to build documentation habits that will hold up when the capability returns. We can do an AI audit-proofing review on your filing this year and make sure you're in good shape whenever enforcement ramps back up."

That framing advances the advisory relationship and builds the service without fearmongering.

---

Three Documentation Practices That Reduce AI Audit Flags

These are not theoretical best practices. They are specific workflow changes that directly address what IRS AI systems are trained to detect.

1. Explain anomalies before filing, not after.

For every return item that represents a statistical departure from historical patterns — a major deduction change, a significant income swing, an unusual one-time transaction — attach a brief explanatory note to the return workpapers. The note doesn't need to be long. It needs to exist. "Client's equipment expense increased due to Section 179 election on vehicle purchase — see attached purchase receipt and business use log." That attachment is the difference between a clean pass and a flag that requires a phone call to resolve.

Make this a default step in your return preparation checklist, not an exception for clients you expect to get audited.

2. Use precise figures wherever actual records exist.

Round numbers on high-variance line items are one of the most consistent AI audit trigger patterns in public IRS guidance. Not because $4,000 is wrong and $3,847 is right — but because precision signals that actual records were consulted, and estimation patterns correlate with underreporting in the training data.

Review any Schedule C, Schedule E, or business return with round-number entries above $500. Where actual figures are available from receipts, bank statements, or invoices, use them. Where clients provided estimates, flag the gap and request documentation before filing.

3. Reconcile every third-party document before the return leaves your desk.

IRS AI automatically cross-references 1099s, K-1s, W-2s, 1098s, and broker statements against filed return data. A mismatch — even a small one — does not get reviewed by a human first. It gets flagged automatically.

Your reconciliation checklist is now a compliance tool, not just a quality control step. Every third-party document your client received should appear on the return in the expected location. Every figure that a third party reported should reconcile to what was filed. Unexplained discrepancies should be addressed in the workpapers before filing — not resolved reactively after an automated inquiry.

---

Frequently Asked Questions

Is the IRS using AI to select tax audits in 2026?

Yes. GAO-26-107522, released March 24, 2026, confirmed the IRS runs 126 active AI use cases — including anomaly detection systems designed to flag returns for audit before human review. The IRS AI audit selection capability is active, though operating at reduced capacity following the loss of 63 AI-skilled employees from the RAAS division. The IRS also handles over 4.8 million taxpayer calls through AI voice systems.

What triggers IRS AI audit selection in 2026?

Documented triggers include: year-over-year income changes without documentary explanation, deduction ratios that are statistically extreme for a given income level and industry, round-number estimates on high-variance line items, self-employment income that doesn't reconcile to third-party reporting, and mismatches between K-1s or 1099s and the filed return. Each of these triggers is addressable through documentation — an explanatory attachment prepared before filing, not after a notice arrives.

Can a CPA firm offer "AI audit-proofing" as a service?

Yes — it's a natural extension of what CPA firms already do. The service is a pre-filing documentation depth review: identify every AI-flaggable pattern in a return, confirm the workpapers explain each one, and remediate any documentation gaps before filing. Price it as a flat-fee add-on ($200–$500 per complex return) or bundle it into an advisory retainer. The positioning: "We review your return the way the IRS AI will."

How did the DOGE staff cuts affect IRS AI enforcement?

The GAO found the IRS lost 63 AI-skilled employees from its Research, Applied Analytics and Statistics group — the division responsible for building and validating AI audit models. The GAO concluded this "increases the risk that IRS AI efforts will not succeed." The practical result: IRS AI-powered audit selection is degraded in 2026 relative to projections, particularly for complex, high-income, and multi-entity returns. The window is temporary. IRS AI capability is expected to rebuild over 12–24 months as the agency restaffs.

What should small CPA firms tell clients about AI audit risk right now?

Three points: (1) IRS AI audit tools are active but operating below projected 2026 capability. (2) Clean documentation and precise reporting remain the correct posture regardless of enforcement intensity. (3) Any client with anomalous return characteristics — income swings, complex deductions, significant self-employment — should have explanatory workpaper attachments prepared at filing. The advice is not "take more risk." It is "use the window to build documentation habits that will hold up when IRS enforcement resumes at full capability."

---

Related Reading

• The IRS Is Now Using AI to Select Audits — What Every CPA Firm Owner Needs to Know — The GAO-26-107522 deep-dive: what the IRS AI infrastructure does and what CPA firms need to adjust
• The IRS Just Lost Its AI Audit Team. Here's What That Means for Your Accounting Clients. — The DOGE staff cut story and what it changes about the enforcement timeline
• AI Compliance Checklist for CPA and Accounting Firms (2026) — AICPA professional standards, Colorado AI Act applicability, and a 5-step compliance checklist
• AI Accounting Task Automation: What Small Firms Should Automate First — The seven tasks ready for automation now, with tool shortlist and ROI math
• AI ROI for Professional Services Firms — Capacity recovery math across firm types and sizes, including advisory service add-ons