(UNITED STATES) — The IRS is using AI and data analytics to help select tax returns for audit, while keeping human review in the process before any audit decision is made.

The agency’s AI governance policy places that work inside its most sensitive category. It says “high-impact” AI includes systems that “inform[] or influence[] whether a taxpayer will be subject to audit, or what aspects of a return will be subject to audit.”

That language places audit selection inside a formal IRS AI governance policy, rather than in a loose or experimental category. It also draws a line between tools that rank risk and a fully automated system that decides, by itself, whether a taxpayer gets audited.

Recent official statements point in the same direction. An IRS AI governance update cited in public reporting said the agency must track high-impact AI when it affects audit selection or audit scope.

Frank Bisignano, identified in April testimony as IRS CEO, also described a broader technology push inside compliance work. The IRS is “undertaking efforts to improve [tax] collections in a manner that employs data, analytics and improved technology to focus compliance efforts where they matter most,” Bisignano said.

Government Accountability Office figures show how large that effort has become. The GAO reported that the IRS had 126 AI use cases, with two-thirds initiated between June 2022 and June 2025, and about a third operational as of March 2024.

Within compliance, the uses described are not abstract. The IRS is using AI to select returns for audit and to detect errors and fraud. That places the technology close to one of the agency’s most sensitive functions, even if it does not make the final call on its own.

The audit-selection process also builds on older analytic systems rather than starting from scratch. The IRS Discriminant Income Function already used “basic artificial intelligence and data analytics” to help select returns for audit, according to the cited material.

That history matters because the current push looks more like an expansion than a sudden break. Newer tools appear to add speed, scale and pattern detection to a process that already relied on data scoring, income analysis and return comparisons.

Work described in 2024 shows the agency examining broader applications. A request for information and later reporting described the IRS exploring AI for real-time return checks and case selection, including models aimed at complex partnerships and high-risk returns.

One contract offers a more concrete look at that experimentation. The IRS paid $1.8 million to Palantir for a case-selection contract testing three AI models, according to the cited material.

Those details suggest a widening role for AI inside enforcement and screening, but they do not support claims that the IRS has handed audit decisions entirely to machines. The material describes AI as a risk-ranking tool, not the sole decider of guilt or fraud, and says human review still stands between a flagged return and an audit decision.

That distinction is central to the current AI governance policy. A system that influences whether a return gets examined can still shape outcomes in a direct way, even if a person signs off at the end. Human review limits full automation, but it does not erase the effect of the score, ranking or recommendation that reaches the reviewer first.

Many of the factors long associated with IRS scrutiny remain familiar in this newer system. The audit triggers most often tied to IRS attention still include high income, complex returns, aggressive credits, unusual charitable deductions, and high business or mortgage-interest deductions.

In practice, that means AI appears to operate less as a replacement for traditional audit markers than as a tool that sorts, prioritizes and compares them across large numbers of filings. A complex return with unusual deductions would already draw interest; AI can help decide how quickly it rises to the top of the pile and what part of the filing receives closer attention.

Questions about bias remain unsettled. Independent concerns about unfair audit selection continue, and past studies showed Black taxpayers were audited at higher rates.

The GAO has also identified “unintentional algorithmic biases” as a possible contributor. That phrasing does not accuse the agency of deliberate discrimination, but it does place automated and semi-automated systems inside a long-running argument over fairness in tax enforcement.

Bias concerns are especially sharp in audit selection because taxpayers rarely see the mechanics behind the decision. If an AI system influences who gets examined, lawmakers and outside watchdogs are likely to press for clearer rules on testing, documentation and review. The IRS AI governance policy already signals that the agency treats such systems as high-impact, which carries its own oversight implications.

Some public claims go further than the confirmed record supports. The confirmed facts are narrower: the IRS is using AI to assist audit selection and other compliance work, the agency classifies that role as high-impact when it influences audit selection or scope, and human review remains part of the process.

What is not supported is the idea that the IRS has fully automated audit decisions. Nothing in the material describes a machine-only system that decides, without human review, who will be audited or what fraud finding will be made.

Several parts of the picture remain incomplete in public. The material does not resolve how far the IRS has gone in reducing bias inside these systems, when real-time return checks might move from exploration to wider use, or how the individual models tested for case selection were designed.

Even so, the direction of travel is plain. The IRS is building AI deeper into compliance work, using policy language that recognizes the stakes and operational efforts that range from broad use-case development to targeted contracts.

That leaves two debates running side by side. One concerns enforcement efficiency, with agency officials describing data, analytics and improved technology as a way to focus compliance efforts. The other concerns fairness, with critics and oversight bodies watching whether AI shifts old disparities into newer technical systems.

Taxpayers are unlikely to see a dramatic change in the basic reasons a return attracts attention. High income, complicated structures, aggressive claims and unusual deductions remain part of the audit picture. What AI changes is the sorting process behind the scenes, where returns can be screened, ranked and routed with far greater speed.

Policymakers now face a familiar problem in a new form: how to permit broader use of AI inside tax administration while keeping audit decisions accountable to human judgment. The current IRS AI governance policy answers part of that by treating audit-related systems as “high-impact.” The harder question is whether that label will lead to safeguards strong enough to address the bias concerns that have followed audit selection for years.