How the IRS Is Using AI to Select Audit Targets and Detect Tax Fraud

The IRS is quietly deploying artificial intelligence — and it's not using ChatGPT the way individuals do. Former IRS Commissioner Danny Werfel describes it as giving IRS personnel "IBM Watson on their side" in a chess match against potential tax evaders: AI that sees patterns across millions of returns that humans would miss.

According to a Government Accountability Office (GAO) report, the IRS has initiated 126 AI use cases — two-thirds of them between June 2022 and June 2025. About 18 of those use cases are specifically focused on tax compliance and fraud detection.

Key Takeaways

• AI-assisted audit selection works via risk scoring. Returns are fed through an AI model trained on historical audit data, which assigns a "risk score" to each. A human review still happens before any audit opens — AI flags, humans decide.
• The IRS is spending tens of millions per year on AI, including contracts with Microsoft, Palantir, and IBM. A $1.8M Palantir contract tested three AI models for case selection alone.
• Staffing cuts are accelerating AI adoption. The IRS has lost nearly a third of its audit staff. AI is partly filling that gap — the GAO report notes that many of the 126 use cases relate to operational efficiency, not just enforcement.

For most taxpayers, the highest audit triggers remain the same: high income, aggressive deductions, complex returns. But as Werfel notes, nobody should underestimate AI's growing power to find what wasn't detectable before.

🔗 Read the full article on South Florida Reporter