The Ethics of Predictive Tax Enforcement
By AG CPA Co. | Perspectives on Finance & Technology | ~630 words
In late 2025, the IRS rolled out a new AI program across its Office of Chief Counsel, Taxpayer Advocate Services, and Office of Appeals. Between January and May of that same year, the agency reduced its workforce by roughly 25 percent — from 103,000 to 77,000 employees. The two facts are related. As headcount falls, technology must absorb more of the enforcement function. The question is not whether AI will play a larger role in tax administration. The question is what ethical constraints should govern how it operates.
What Predictive Enforcement Looks Like
The IRS has been using AI in enforcement contexts for several years. As of early 2024, the Treasury Inspector General for Tax Administration reported 68 active AI-related projects, including 27 specifically focused on compliance and enforcement. These systems use machine learning to analyze millions of returns simultaneously, scoring them for audit potential based on patterns in historical audit outcomes, taxpayer profiles, and external data sources. The IRS has stated it now routinely incorporates data from financial institutions, public records, and social media to corroborate or challenge taxpayer disclosures.
The Holland & Knight law firm, in a November 2025 analysis, described the current IRS approach: predictive analytics flag returns with a higher probability of underreporting or aggressive tax positions; targeted campaigns focus on areas with high audit potential such as personal use of business jets, international asset holdings, and complex partnership structures. For high-income and internationally connected taxpayers, this is not a theoretical concern.
The Bias Problem
The Government Accountability Office has issued multiple reports raising concerns about algorithmic bias in IRS audit selection. The underlying issue is structural: AI systems trained on historical audit data will reproduce whatever patterns exist in that history. If certain industries, geographic areas, income levels, or demographic groups were audited at higher rates in the past — for reasons that may or may not reflect actual noncompliance — a model trained on that data will continue to flag them at higher rates.
A 2025 peer-reviewed study published in Humanities and Social Sciences Communications examined AI in tax administration across multiple jurisdictions and found that the Australian Tax Office had completed data ethics assessments for only 26 percent of its AI models in production. The study also noted that while human oversight remained formally in place for adverse tax decisions, the ATO had acknowledged a declining ability among staff to explain AI-driven outcomes — a troubling signal for any system in which meaningful human review is supposed to be a safeguard.
Transparency and Due Process
In 2022, France's Conseil d'État — the country's highest administrative court — issued a 360-page report on AI in the public sector recommending a doctrine of 'trusted public AI,' grounded in seven principles: transparency, accountability, human oversight, and auditability among them. The United Kingdom's 2025 Elsbury tribunal case, which required HMRC to disclose its use of generative AI in taxpayer correspondence, brought similar concerns into the courtroom for the first time.
The due process stakes are real. A taxpayer who receives an audit notice generated or prioritized by an algorithm they cannot see, based on criteria that are not disclosed, has limited ability to understand whether the selection was legitimate. The EU AI Act imposes transparency obligations on systems that interact with individuals or generate content capable of misleading users — a standard that, applied to tax enforcement, would require significant changes to current practice in most jurisdictions.
What Responsible Enforcement Requires
The Bipartisan Policy Center's 2024 analysis of AI use in tax administration identified several principles for responsible deployment: clear privacy safeguards, codes of conduct, audit trails, and mechanisms for independent oversight. Capitol Technology University researchers, writing in early 2026, proposed establishing a data integrity and ethics lab and bringing in independent auditors to review IRS AI systems — structural safeguards that would create accountability without eliminating the efficiency gains the agency is pursuing.
For practitioners advising clients who face IRS scrutiny, understanding that audit selection is increasingly algorithmic — and knowing what that means for how responses should be structured — is becoming part of the professional's core competency. The intersection of tax law and machine learning is no longer a niche interest. It is where the practice is heading.
REFERENCES
EisnerAmper. 'AI at the IRS: Transforming Tax Enforcement and the Future of Taxpayer Services.' January 2026.
Holland & Knight. 'IRS Audits and the Emerging Role of AI in Enforcement.' November 2025.
Nature / Humanities and Social Sciences Communications. 'Balancing Innovation and Integrity: AI in Tax Administration and Taxpayer Rights.' November 2025.
Capitol Technology University. 'Audited by an Algorithm: How the IRS Is Using AI in 2026.' February 2026.
Bipartisan Policy Center. 'AI Use in Tax Administration.' September 2024.