Audited by an Algorithm: How the IRS Is Using AI in 2026

 

Tax season is here, but this year, the IRS will rely less on human oversight and more on artificial intelligence. Between January and May 2025, the IRS cut its workforce by 25%, from 103,000 to 77,000 employees. As a result, the IRS is leaning more on technology to fill the ever-growing tax gap, with AI now powering audit selection, fraud detection, taxpayer services, and more. While this transformation is improving efficiency, accuracy, and faster processing, it raises critical questions about fairness, privacy, and algorithmic biases.

How AI Affects Tax Season

A variety of machine learning models now analyze millions of tax returns simultaneously, scoring them for audit potential. The tax gap is a growing issue that the IRS hopes to solve. The “IRS’s most recent estimate of the tax gap puts the amount owed and not paid at about $496 billion each year (for 2014-2016),” which is a staggering deficit and workload for IRS staff.

By using AI scoring systems like the Discriminant Function, discrepancies between income and deductions can be flagged and addressed. The Large Partnership Compliance Model analyzes complex partnerships like hedge funds, private equity, and real estate operations that were previously too difficult to audit. In 2021, this model selected 82 high-risk returns compared to only single digits before. For corporations with $10-250 million in assets, the Line Anomaly Recommender has replaced outdated systems.

The Individual Taxpayer Model recommends the top three issues likely needing adjustment on each return. These AI systems run six times per tax year, learning with each iteration. Common triggers include year-over-year income discrepancies, extreme deduction ratios, round numbers suggesting estimates, and underreported self-employment income. AI analyzes patterns across your entire tax history, not just individual line items, looking for unusual deviations from prior filing patterns and flagging returns for review, such as those that show significant year-over-year income underreporting.

With these systems, AI has dramatically improved processing speed. Voice bots have also handled over 4.8 million calls, while chatbots resolved more than 450,000 inquiries with 42% of users never speaking to a human agent. In late 2025, the IRS deployed the Agentforce system across multiple divisions to summarize cases and search documents instantly, lending to efficiency and improved operations.

Issues with Using AI for Tax Return Review

There are notable downsides, however, to moving to AI for the tax return review process. One issue involves AI biases. Independent studies show that Black taxpayers are audited at rates three to five times higher than others. The Government Accountability Office (GAO) identified unintentional algorithmic biases as a potential cause. When AI trains on historical data containing existing biases, it perpetuates past discrimination through automated systems.

With workforce cuts leaving insufficient human oversight, the risk of unfair outcomes grows. Proposed solutions include establishing a data integrity and ethics lab and bringing in independent auditors. Experts point to the Robodebt scheme, an unlawful method of automated debt assessment and recovery implemented in Australia, as a cautionary tale of unchecked algorithms.

There are also major privacy concerns related to using AI to process vast amounts of sensitive financial and personal data. Centralizing and automating this analysis could increase the risk of data breaches or unauthorized access if cybersecurity safeguards fail. Additionally, opaque algorithms may make it difficult for taxpayers to understand how their data is being used and stored, or to challenge decisions influenced by automated systems.

Other issues stem from fewer opportunities to speak with live agents, where routine questions will get automated responses, which can be frustrating to taxpayers. If the AI system isn’t fine-tuned and accurate–or if a person has a complex question outside the algorithm’s script–then these filers who need nuanced guidance may be out of luck.

Transparency and Communication Gaps

The IRS has provided minimal public information about how their algorithms work, citing concerns that taxpayers might hack or “game the system.” The agency previously published specifics, but now it maintains secrecy under law enforcement exemptions.

The GAO has called for better documentation and transparency around the IRS’ use of AI. Advisory panels have recommended improvements, but taxpayers remain largely in the dark about the algorithms making decisions about their returns, and taxpayers selected for audit aren't told whether it was humans or AI that flagged their return.

Artificial Intelligence at Capitol Tech

The IRS now operates 129 AI use cases, up from 54 in 2024. This expansion is creating high demand for AI engineers, data scientists, machine learning specialists, and ethics auditors. Capitol Technology University's BS in Artificial Intelligence prepares you for these roles through rigorous curriculum and our virtual Capitol Artificial Intelligence Lab (CAILIE). Covering AI ethics, machine learning, and human-technology interaction, our program develops AI engineers, data scientists, prompt engineers, and more who can contribute across a wide variety of fields of study and organizations. 

Explore what a degree from Capitol Tech can do for you! To learn more, contact our Admissions team or request more information.

 

Written by Jordan Ford 
Edited by Erica Decker