Why It Matters
Michigan is deploying artificial intelligence tools to process Supplemental Nutrition Assistance Program applications, a move that could affect hundreds of thousands of residents who rely on the benefits for food security. The state’s use of AI in SNAP eligibility reviews comes as federal legislation threatens to penalize states financially for payment errors, raising questions about whether automated systems could create new barriers for vulnerable residents.
Attorneys and advocates familiar with Michigan’s history of automated benefits processing are urging caution, pointing to a widely documented failure in the state’s unemployment system as a cautionary example of what can go wrong when government agencies move too quickly toward automation.
What Happened
The Michigan Department of Health and Human Services has begun using an AI case reading tool to help employees review SNAP cases line-by-line before payments are issued, department Chief Operating Officer David Knezek told members of the Senate Appropriations Subcommittee on DHHS on March 17, 2026.
Knezek said the AI tool allows the department to scan every active case in what he described as “a perfect environment” before money goes out, and to flag cases with the highest likelihood of resulting in a payment error. He told the committee that without AI assistance, the department is only capable of manually reviewing a relatively small number of cases.
The department is also deploying an optical character recognition tool to scan and digitally input documents submitted by applicants, such as pay stubs, reducing human data entry errors on the front end while retaining human verification on the back end. Knezek said the largest number of errors occur in single and dual-person households, while the largest dollar-value errors tend to come from larger households.
The Federal Pressure Behind the Move
Michigan’s acceleration of AI tools is directly linked to new requirements under H.R. 1, known as the One Big Beautiful Bill Act. Under the legislation, states would be required to pay for a portion of SNAP benefits out of their own budgets based on their payment error rate — a metric measuring how accurately states determine eligibility and benefit amounts.
The nonpartisan Brookings Institution has noted that under the measure, wrongly rejecting an eligible applicant is not counted as an error, meaning the error rate is not a comprehensive measure of accuracy or fraud. Critics argue this structure creates financial incentives for states to focus on overpayments while potentially overlooking wrongful denials.
By the Numbers
- Michigan DHHS serves approximately 1.4 million residents through the SNAP program, according to recent state estimates.
- Payment error rates in SNAP nationally have hovered between 6 and 11 percent in recent federal reporting years, according to USDA data.
- Under H.R. 1, states with error rates above a federal threshold could face significant cost-sharing obligations that currently fall entirely on the federal government.
- The Brookings Institution estimates that millions of eligible households nationwide could face disruptions if states implement aggressive error-reduction strategies that conflate errors with fraud.
- Michigan’s MiDAS unemployment fraud system, deployed in 2013, wrongly accused an estimated 40,000 residents of fraud before the state acknowledged system failures.
Zoom Out
Michigan’s move reflects a national trend of states integrating AI and automated decision-making tools into public benefits administration, often driven by cost pressures and federal compliance requirements. Several other states, including Arkansas and Idaho, have experimented with algorithm-driven eligibility screening in Medicaid and SNAP programs, with mixed results and ongoing litigation in some cases.
The concern raised by Michigan attorneys centers on the state’s own history with the Michigan Integrated Data Automated System, or MiDAS, which was used to process unemployment insurance claims. That system wrongly flagged tens of thousands of residents as unemployment fraud suspects between 2013 and 2015, triggering automatic penalties before human review occurred. The state eventually paid out tens of millions of dollars in settlements and restitution after courts found the automated system had violated due process rights.
Advocates say the MiDAS episode demonstrates that AI and automated tools in benefits processing require robust human oversight, transparent audit mechanisms, and accessible appeals processes to prevent wrongful denials or penalties at scale.
What’s Next
The Michigan Advance has submitted questions to MDHHS regarding the specific deployment dates of both the AI case reading tool and the optical character recognition system, as well as details about vendor contracts and oversight protocols. Those responses had not been provided as of publication.
Legislative review of the department’s AI deployment is expected to continue through the Senate Appropriations Subcommittee on DHHS. Advocates and legal organizations are anticipated to push for public transparency requirements and formal error-appeal processes as the tools expand across the state’s SNAP caseload.