ACLU Questions AI Use in Michigan’s SNAP Payment Accuracy Checks

Amid growing concerns about the reliability of artificial intelligence in public services, the ACLU of Michigan is seeking transparency on the state’s use of AI to regulate SNAP benefit payments. The organization’s request for records aims to scrutinize the technology’s accuracy and ensure it doesn’t mistakenly withhold rightful benefits.

Last week, the American Civil Liberties Union (ACLU) of Michigan submitted a formal request for information regarding the artificial intelligence system employed by the state’s health and human services department. This AI tool is tasked with ensuring the accuracy of Supplemental Nutrition Assistance Program (SNAP) benefit payments, a move that has raised eyebrows due to known issues of AI errors.

The ACLU’s attention was drawn to this AI deployment following testimony from David Knezek, Chief Operating Officer of the Michigan Department of Health and Human Services (MDHHS), during a State Senate subcommittee meeting in March. Knezek revealed that the department had introduced an AI tool to audit SNAP payments before distribution statewide. Watch the testimony here.

“We are doing a pre-certification enabled case read,” Knezek stated, highlighting the AI tool’s capacity to process a high volume of payments. “We’ve done 17,500 of them just in the last three months,” he noted, emphasizing the technology’s efficiency compared to manual reviews.

Knezek further disclosed that the AI system identified over $150,000 in potential payment discrepancies from around 1,500 case reads.

ACLU’s staff attorney Phil Mayer sent a letter to MDDHS requesting detailed information about the AI’s operations and the measures in place to prevent wrongful denial of benefits. Mayer expressed a desire to understand the AI’s effectiveness and its handling of complex applications, especially those from non-native English speakers. Read the full letter here.

“We want to understand what evidence they have that it works properly,” Mayer 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.”

The backdrop to these concerns is the One Big Beautiful Bill Act, which mandates that states with SNAP error rates above 6% must reimburse a portion of their program costs. Michigan’s error rate last year approached 10%, leading to financial implications for the state. See more on state payment error rates.

Moreover, past experiences with automated systems in Michigan have been problematic. In 2024, the state settled a $20 million lawsuit related to the Michigan Integrated Data Automated System (MiDAS), which wrongly accused thousands of residents of unemployment fraud. This incident led to severe financial distress for many, with over 1,000 individuals declaring bankruptcy. Read about the settlement. Details on the MiDAS issue. Impact on residents.