The ACLU of Michigan put in a request last week for records relating to an artificial intelligence program used by the state's health and human services agency to make sure that SNAP benefit payments were accurate.
The civil rights organization raised concerns about tasking AI with rooting out erroneous food assistance payments, since the technology has been known to make errors.
The ACLU became aware of the use of AI by the Michigan Department of Health and Human Services through testimony by the agency’s chief operating officer, David Knezek to a subcommittee of the State Senate in March. He said that the agency had recently launched an artificial intelligence tool to review SNAP payments before they were issued across the state.
“We are doing a pre-certification enabled case read,” Knezek said, and added that the AI tool had made it possible to review many more payments.
“We've done 17,500 of them just in the last three months," Knezek said. “This is a massive improvement using an AI tool over what can be done by an individual person or an individual team over the course of a year.”
In about 1,500 case reads, he said, the AI tool had “caught more than $150,000 worth of potential payment errors.”
Phil Mayer, a staff attorney at the ACLU of Michigan, wrote a letter to MDDHS asking for more clarity on the role of AI and safeguards to ensure that payments that have been properly cleared are not being erroneously withheld.
“We want to understand what evidence they have that it works properly,” he said. “We want to understand whether it makes mistakes particularly when reviewing complicated applications or second-language speaker applications and we want to know if they complied with the process of securing federal approval before moving forward with the technology.
"We have a real concern when … artificial intelligence is used to make decisions about benefits, particularly where there's incentives to deny the benefits in error as here.”
As part of the One Big Beautiful Bill Act — President Donald Trump's signature tax-and-spending legislation — states with high error rates will be required to pay a portion of SNAP benefits. As per the law, states with error rates higher than 6% will be responsible for covering a portion of their states’ costs — up to 15% — for the program, which is usually federally funded.
Last year, Michigan’s error rate was nearly 10 percent.
In 2024, a different state agency finalized a $20 million settlement after wrongly accusing 3,000 Michiganders with unemployment fraud. The Michigan Integrated Data Automated System (MiDAS) had sent out automated messages to tens of thousands of Michigan residents citing them for wrongfully taking payments from the state’s unemployment benefits program between 2013 and 2015.
More than 1,000 of them declared bankruptcy because they were unable to pay back the amount MiDAS claimed they owed — along with repayment fines and 400% interest.
