What does an AI-driven IRS mean for small accounting firms?

The IRS now uses 126 active AI projects, up from just 10 two years ago. This rapid change, which has not been widely communicated to professionals, is changing how the IRS selects audits, detects fraud, and enforces rules. These changes directly impact the clients of small accounting firms every day. 

The IRS has fundamentally changed how it enforces tax compliance. As of June 2025, the agency operates 126 active AI use cases, up from just 10 in August 2022, spanning audit selection, fraud detection, identity verification and compliance scoring. This is not a modernization initiative still in the planning stage. It is an operational reality that is already determining which returns get flagged, which partnerships get examined, and which CP2000 notices land in your clients' mailboxes.

For small accounting firms, the shift matters because the IRS against which your clients filed last year is not the same agency reviewing their returns today. The selection criteria have changed. The data matching has changed. The enforcement priorities have changed. And the workforce executing all of it has shrunk by 25%, leaving AI to fill the gap.

According to the TIGTA FY2026 Management and Performance Challenges Report published in October 2025, the IRS reported 101 active AI projects as of April 2025, up from 68 just one year earlier, with 27 of those specifically focused on tax compliance and enforcement. The GAO's March 2026 report (GAO-26-107522) documented 126 use cases as of June 2025, categorized across operational efficiency, customer service and, most critically for practitioners, tax compliance and fraud detection.

The enforcement infrastructure has moved from pilot programs to production deployment. Practitioners who treat this as a future concern are already operating with outdated assumptions.

What AI is actually doing inside the IRS right now

The IRS is using AI across three core enforcement functions: audit case selection, third-party data cross-matching and fraud detection. Each function directly affects how small accounting firm clients are evaluated before, during and after the filing process. Understanding what AI is doing inside the IRS is the first step toward advising clients effectively in this environment.

Audit selection: from statistical scoring to machine learning

The IRS has historically used the Discriminant Information Function, a statistical scoring model, to rank returns by audit potential. That DIF model produced notoriously high "no-change" audit rates, meaning the IRS was selecting and opening cases that yielded little or no additional tax. AI is replacing that model with something significantly more precise.

According to a GAO analysis of IRS AI deployment, the IRS now operates two separate AI models to prioritize large partnership returns for examination. For individual returns, AI models select a stratified sample and identify those statistically most likely to contain errors or underreported income, with early pilot results showing the new models outperform prior selection methods by a meaningful margin.

The pressure driving this investment is quantifiable. The gross tax gap stands at $688 billion for tax year 2021 and $696 billion for tax year 2022, the difference between taxes legally owed and taxes actually paid. With Congress reducing IRS funding from $80 billion over 10 years to $37.6 billion, and a proposed FY2026 budget cutting annual funding by a further 20%, AI-driven enforcement efficiency is no longer optional for the agency. It's the only path to closing the gap with fewer resources.

Three taxpayer categories are explicitly in the IRS's AI enforcement crosshairs. A TIGTA report from May 2025 (No. 2025-308-022) confirmed the IRS Small Business/Self-Employed Division has built a dedicated AI return selection model that applies individualized risk ratings to self-employed individual income tax returns. The three categories: large partnerships, self-employed taxpayers, and small business owners. For most small accounting firms, this covers a substantial portion of the active client roster.

Fraud detection and automated cross-data matching

The IRS's AI-driven fraud detection operates by ingesting third-party data, W-2s, 1099-NEC, 1099-K, 1099-INT, FBAR and FATCA filings, and cross-referencing them against filed returns at machine speed. Discrepancies that would have taken a human reviewer days or weeks to identify are now flagged in real time, generating automated notices before a practitioner has any opportunity to intervene.

The most significant new data source is Form 1099-DA, which went into effect for the 2025 tax year. Under the IRS's final regulations, brokers are now required to report gross proceeds from digital asset transactions directly to the IRS, creating a third-party reporting trail for crypto activity that previously did not exist. The IRS uses AI to auto-match those reported proceeds against filed returns. Any mismatch triggers an automated CP2000 notice or referral for examination.

The critical detail for practitioners: per IRS guidance FS-2025-06, taxpayers are responsible for calculating their own cost basis. The broker reports gross proceeds only. If a client's basis calculation is missing, estimated or inconsistent with the gross proceeds reported, the IRS's AI will flag it, and the notice will arrive before you know there is a problem.

Beyond digital assets, the IRS AI system flags specific patterns across all return types: low taxable income relative to indicators of wealth or asset accumulation; multi-year discrepancies where reported income or balance sheets do not reconcile year-over-year; round numbers in deduction schedules that suggest estimation rather than actual records; and partnership allocations that appear economically inconsistent. These are not hypothetical risk factors. They are documented in GAO's published analysis of IRS AI red flags and should be treated as the effective compliance standard for every client return.

On Feb. 10, 2026, the IRS codified its AI enforcement practices into formal policy. IRM 10.24.1, the first official Internal Revenue Manual section governing the use of AI in audit selection and exam support, establishes three things: what AI is authorized to do in the examination process, how mandatory human review is integrated into AI-assisted case selection, and what documentation requirements apply to AI-generated referrals.

The significance of IRM 10.24.1 for practitioners is this: It removes any remaining ambiguity about whether AI is experimental or operational inside the IRS. It is now a codified policy. The IRS has institutionalized AI in its enforcement workflow, with human review structured around AI outputs rather than the other way around.

The IRS is deploying more AI at the same time it is operating with significantly fewer people. That combination, more automated enforcement capability, and a dramatically reduced human capacity, creates a specific and immediate operational problem for small accounting firms: Notices and flags will arrive faster, but resolution will take longer.

The scale of IRS workforce reduction

The numbers are documented and significant. According to the TIGTA FY2026 Management and Performance Challenges Report, the IRS lost approximately 25% of its total workforce between January and May 2025, shrinking from roughly 103,000 to approximately 77,000 employees. The reductions were not evenly distributed:

• The IT department lost 25% of its workforce, the same team responsible for maintaining and developing the AI systems now central to enforcement.
• Return Integrity and Compliance Services, the division responsible for detecting fraudulent refunds, lost 18% of its staff. TIGTA estimates this reduction alone could result in approximately $360 million in fraudulent refunds going undetected in the next filing season.
• Revenue agents, the staff responsible for complex examinations of large corporations, partnerships, and high-wealth individuals, declined by 31% in the same period, according to research from the Budget Lab at Yale.
• The Research, Applied Analytics, and Statistics group, the internal team supporting AI development and governance, lost 63 employees who had been working full- or part-time on AI, per the GAO report.

Ongoing audits and examinations were dropped

The workforce reduction not only affected future enforcement capacity, it also affected cases already in progress. The TIGTA report confirmed that not all large partnership returns were analyzed for audit selection due to resource constraints, and the dedicated IRS unit formed to audit large pass-through entities was largely dismantled.

As a direct consequence, a number of ongoing audits and examinations, particularly of large corporations and partnerships, were closed or abandoned midstream. Former IRS agents reported to the International Consortium of Investigative Journalists that active corporate audits were dropped due to staffing shortfalls, potentially leaving significant uncollected revenue on the table. Skadden's January 2026 analysis confirmed the IRS's Global High Wealth Program and pass-through entity enforcement efforts were largely hollowed out during this period.

The net effect for small accounting firms

AI is identifying compliance issues faster than at any point in the IRS's history. The human infrastructure to resolve those issues has simultaneously contracted by 25%. For small accounting firms, this creates an asymmetric problem: Automated flags and notices arrive quickly, but reaching a human agent, obtaining a substantive response, and navigating a notice to resolution now take significantly longer than they did two years ago. Firms need to build notice-response capacity into their standard workflows and set realistic expectations with clients before issues arise, not after.

The AI-driven IRS changes the compliance risk profile of the clients that small accounting firms serve. The compliance bar has risen, the documentation standard has shifted, and the advisory value firms can provide has expanded, but only for firms that recognize the change and position accordingly.

The IRS does not publish the criteria its AI models use to select returns for examination. GAO has explicitly flagged this as a governance failure: per the GAO report, the IRS has not established sufficient internal controls to ensure its AI models are consistently documented, independently validated, or explainable to affected taxpayers. Practitioners have no official disclosure mechanism to determine what is triggering scrutiny on any given return.

For SMB-serving firms, this opacity directly impacts clients. Consider the typical client roster: S corps with mixed income streams across multiple entities; real estate operators with depreciation-heavy Schedule E returns and year-over-year balance sheet fluctuations; self-employed contractors with variable income and home office deductions; ecommerce operators with high gross revenue and compressed margins. 

Each of these profiles contains the types of patterns, income variability, asset-income ratios and deduction clustering that AI models are specifically trained to identify as elevated risk. The compliance bar has risen, but the IRS has not told practitioners where the new bar is.

Documentation standards have effectively increased

AI-driven audit selection is more sensitive to anomalies, inconsistencies, and statistical outliers than the manual review it replaces. GAO-24-106449 explicitly recommends that the IRS improve documentation of its AI models to ensure consistency and transparency, but until that governance improvement is implemented, practitioners must assume that AI is flagging anomalies on returns without providing any explanation of why.

The practical documentation standard has changed as a result. What was sufficient to satisfy a human reviewer — a general description of business purpose, a ballpark mileage estimate, and a verbal understanding with a client about mixed-use assets — may not survive an algorithm trained to detect statistical outliers across hundreds of millions of returns. 

Practitioners need to raise the documentation bar with clients before filing, including contemporaneous records for deductions, written business purpose statements for mixed-use expenses, and consistent year-over-year treatment of recurring line items. AI flags the anomaly; strong documentation defends against it.

The advisory opportunity is real and expanding

As AI handles more of the transactional compliance work, data cross-referencing, form matching and anomaly detection, the function that AI cannot perform is interpretation. AI can flag a partnership allocation that appears statistically inconsistent. It cannot explain the economic rationale behind it, prepare the client for an IRS inquiry, or advise on restructuring to reduce future risk. That is the practitioner's role, and it is expanding in direct proportion to what AI is automating.

Small accounting firms that reframe their client relationships from annual return preparation to year-round compliance risk management are better positioned in this environment for two reasons. First, the compliance risk is now continuous, not confined to the filing season. AI is running cross-matches and flagging anomalies year-round. Second, the advisory conversation has a clear, data-supported foundation: The GAO and TIGTA have publicly documented the IRS's AI targeting priorities, the known red flags and the enforcement gaps. That documentation gives practitioners a credible basis for proactive client guidance that goes beyond generic tax advice.

Response timelines are being tested

AI-generated CP2000 notices and examination referrals are issued at machine speed. The IRS workforce available to respond to practitioner inquiries, process responses and close cases has contracted by 25%. The gap between how fast the IRS can initiate contact and how fast it can resolve it has widened considerably.

Practitioners are reporting longer resolution cycles across routine notice types. Power of Attorney processing times have extended. Examination assignment timelines are less predictable. Firms that do not build structured notice-response workflows, intake, triage, POA filing and substantive response tracking are likely to experience client service failures as the volume of AI-generated notices increases, as projected. Managing client expectations about IRS resolution timelines is now a core component of client service, not an exception.

How small firms should respond: 5 practical steps

Small accounting firms need to make five specific adjustments to operate effectively in the AI-driven IRS environment. 

These are not aspirational recommendations; they are practical workflow and advisory changes grounded in what GAO, TIGTA, and the IRS's own published guidance tell us about how AI enforcement is currently operating.

Step 1: Build contemporaneous documentation into every client engagement. Do not wait for a notice to assess whether a client's records are sufficient to defend their return. AI flags the anomaly; documentation resolves it. For every client with non-standard deductions, mixed-use asset claims, year-over-year income variability or complex entity structures, establish a documentation standard before filing. Automated bookkeeping services can help maintain real-time expense logs, written business purpose statements for deductions, and year-over-year consistency in how recurring items are treated and reported. The investment is minimal upfront. The cost of assembling documentation after an AI-generated notice arrives is substantially higher.

Step 2: Review returns against documented IRS AI red flags before filing. Per the GAO's published analysis of IRS AI deployment, the IRS's AI models are specifically trained to identify the following risk patterns: taxable income that is low relative to indicators of wealth or asset accumulation; multiyear discrepancies where reported income or balance sheet figures do not reconcile year-over-year; round numbers in deduction schedules that suggest estimation rather than actual record-keeping; and partnership allocations that appear economically inconsistent with the structure of the entity. 

Add a pre-filing review step to your workflow that checks every client return against these documented criteria. No new software is required; a disciplined checklist built from GAO and TIGTA's published findings is sufficient. The review adds minimal time and meaningfully reduces client exposure.

Step 3: Make Form 1099-DA basis reconciliation a standard pre-filing step. For any client with digital asset activity in tax year 2025, basis reconciliation is now mandatory pre-filing, not optional. Under IRS final regulations and FS-2025-06 guidance, brokers are reporting gross proceeds from digital asset transactions to the IRS. The IRS's AI cross-matches those proceeds against filed returns. 

Taxpayers calculate their own basis. Any gap between reported proceeds and the client's basis calculation is an immediate AI flag. If your client holds or transacted crypto in 2025, that conversation must happen before the return is filed, not after the CP2000 arrives.

Step 4: Implement a structured notice-response workflow. Create a clear process for handling IRS notices. Due to a 25% cut in IRS staff, it is taking longer to resolve all types of notices. The process should include: a specific way to receive and file any IRS correspondence clients receive; filing a Power of Attorney right away when needed; a response schedule with internal deadlines; and a way to communicate with clients that sets realistic expectations about how long IRS responses will take. Firms that react to notices as they come in, without a clear process, will create larger backlogs as the number of AI-generated notices increases.

Step 5: Reposition the firm as a year-round compliance risk partner. The strongest competitive advantage for small accounting firms in the AI-driven IRS environment comes from relationships, not technology. AI can identify issues in tax returns, but it cannot explain the context behind those issues, help clients prepare for audits or provide advice on structuring decisions to minimize future compliance risks. 

Small firms can have these important conversations, but only if they stay engaged with clients year-round instead of just during tax season. Sharing updates on IRS AI enforcement, reviewing client books every few months for potential risks and being the first call clients make when they receive an IRS notice are all ways to build a strong, compliance-focused partnership that goes beyond standard tax preparation.

The bigger picture: uncertainty is the operating environment

The IRS's AI transformation is real, accelerating and structurally incomplete, all three at once. GAO-26-107522 found that 61% of the IRS's 126 AI use cases were still listed as "in development" as of June 2025. The agency has not developed a workforce plan to identify the skills needed to support its AI initiatives, a gap the GAO explicitly flagged as a long-term risk to program success. The IRS has also not implemented sufficient internal governance to ensure its AI models are consistently validated, documented, or explainable, meaning the models currently in production are operating without the oversight framework that the GAO has recommended.

At the same time, the agency's own FY2027 Congressional Budget Justification states plainly: "Without modernization, the IRS would be unable to sustain performance with a reduced headcount." The IRS has made a structural bet that AI can compensate for fewer people. Whether that bet fully succeeds is uncertain. What is not uncertain is that the enforcement infrastructure currently live, audit selection models, 1099-DA cross-matching and IRM 10.24.1 governance are operating now and affecting returns being filed today.

The firms that wait for the IRS's AI program to stabilize before adjusting their practice will be the ones most exposed when the next wave of AI-generated notices reaches their clients.

The AI-driven IRS is not an emerging trend. The current compliance environment affects every aspect of small-business tax filing, and it rewards practitioners who understand it over those who are still operating on assumptions formed before 2025. 

For small accounting firms, the core implications are clear and immediate: The effective compliance standard has risen without public disclosure of where it now sits; documentation requirements have increased without formal guidance saying so; notice volumes will grow as AI cross-matching expands; and human IRS capacity to resolve those notices has contracted by 25%.

The firms best positioned to thrive are not necessarily the largest or the most technologically sophisticated. They are the ones that understand what the IRS's AI is actually doing, translate that understanding into proactive client guidance, and use this shift as the foundation for deeper, year-round advisory relationships that no algorithm can replicate.