Press Releases

Today, Congresswomen Suzan DelBene (WA-01) and Gwen Moore (WI-04), along with Members of the House Ways and Means Committee, sent a letter to the Internal Revenue Service (IRS) regarding the agency’s use of Artificial Intelligence (AI) to write code used to manage Americans’ taxpayer data.

A recent study found that AI-generated source code is frequently inaccurate. Developers often spend as much or more time checking and rewriting the code that AI systems produced than if they were to write it themselves.

The lawmakers are seeking answers about the IRS’s plan to rely on AI tools to interface with its decades-old Common Business-Oriented Language (COBOL) systems, which continue to underpin the Individual Master File (IMF), the core repository for taxpayer information since the 1960s. These questions come at an especially critical time when the IRS has already lost 25% of its IT staff since early 2025, and the next tax filing season is just around the corner.

“We hope this Administration shares our goal of ensuring any use of AI by the IRS is appropriately scrutinized, especially given the importance and significance of maintaining the integrity of the IMF and protecting taxpayer data from inappropriate access or exposure,” the lawmakers wrote.

They expressed concerns about the risks of using AI, especially when dealing with systems that manage sensitive taxpayer information and which are also responsible for processing tax returns for hundreds of millions of Americans. In the letter, the lawmakers sought information on the agency’s current and future AI use with its legacy COBOL systems, efforts to address potential inaccuracies and limitations, and taxpayer data protections.

Questions raised include:

• To what extent has the IRS already started deploying AI to interact with its legacy COBOL systems and what future deployments are planned? 
• What processes are in place to review the conversions and AI source coding before, during, and after new AI-created code implementation to ensure it is accurate, including ensuring that AI-generated code is free of hidden errors or unintended behaviors that might not appear during standard testing?
• What specific protections for the taxpayer data protected by Internal Revenue Code section 6103 are in place?

The letter requested a written response from the IRS no later than January 10, 2025.

Read the full text of the letter here.