IRS eyes agentic AI tool to help staffers process potential fraud cases

The IRS is piloting an agentic AI tool that will handle “the rote” steps of addressing potential fraud cases, an agency official said Friday.

Jim Clifford, director of return integrity and compliance services in the IRS’s Taxpayer Service Division, said during a panel discussion at ACT-IAC’s Service Delivery and Experience Summit that the agency is exploring how to use agentic AI after anomalies in tax returns are flagged.

The IRS currently uses machine-learning models to identify anomalies. Returns containing those anomalies are then routed into “buckets, for lack of a better technical word,” Clifford said, where they’re compared against “a whole bunch of other data” to determine if there’s a high risk of identity theft or fraud.

The process of “detecting fraud at the front door” is something Clifford believes the IRS is “pretty good at.” But the “back end” is where agentic AI would come into play, he said, helping agency employees be “more efficient” by handling the “rote steps of an audit or a fraud detection or a fraud remediation case.”

In a sideline interview after the panel, Clifford told FedScoop that the “clerical, rote work” he was referring to includes organizing correspondence with the taxpayer and reading through the pages that person sends to the IRS. 

“What we want the agentic AI to do is review the response, summarize it, build a case, and hand it off to an employee at the point where it’s really ready to make a decision,” he said. 

The distinction between AI detecting fraud and AI helping expedite the process of resolving fraud cases is key: Clifford said the technology addresses the “constraints” agency workers have on how fast they can work, but “making judgments” will always be under a human’s purview.

Clifford didn’t have a timeline for deployment of the agentic AI tool, but he said the pilot is being developed “on another workstream” and will eventually be scaled to other areas of the IRS “once it’s tested and proven to be successful.” 

Any edge the IRS could get to combat fraud would be a welcome development given the stakes. According to the Government Accountability Office, the federal government loses between $116 billion and $304 billion annually to tax fraud.

Every month, the IRS faces approximately 1 billion attempts from fraudsters trying to obtain unauthorized information, Clifford said, largely because “your tax returns are gold” to them. The agency is “pretty damn good at protecting” taxpayer information, though there are “very few times” in which information is “released inappropriately,” he added.

Even though the IRS is “99% accurate” in thwarting attempted identity theft refund fraud “based on all estimates and studies,” Clifford said the cost of fraud is still too much; “substantial” doesn’t go far enough.

Beyond the likely agentic AI assist the IRS has coming, Clifford said more data-sharing with agency and state and local governments could help in the fight, as could more “interoperability” with its partners. 

“There are legislative and regulatory constraints about us sharing data across agencies, but there are really good things we could do if we could do that,” Clifford said on stage. 

“We have the technology now with APIs, with the use of” AI, he continued. “We can share information in a very safe way across agencies to serve the American taxpayer better and to protect the American taxpayer better. And I’d love to see us get that breakthrough.”