Florida may hire an AI vendor to help decide which hungry families get grocery assistance
The new state budget includes a $4 million contract for a vendor that can use machine learning to make eligibility decisions for an anti-hunger program that feeds nearly 2 million Florida families.
This is Seeking Rents, a newsletter and podcast devoted to producing original journalism — and lifting up the work of others — about Florida politics, with an emphasis on the ways that big businesses and other special interests influence public policy in the state. Seeking Rents is produced by veteran investigative journalist Jason Garcia, and it is free to all. But please consider a voluntary paid subscription, if you can afford one, to help support our work. And check out our video channel, too.
The state of Florida may start using artificial intelligence to decide who gets grocery assistance.
A last-minute provision inserted into the new state budget orders the state’s family-welfare agency to hire an outside vendor that can use machine learning to make eligibility decisions for the Supplemental Nutrition Assistance Program — the critical anti-hunger program that helps feed nearly 3 million Floridians.
Republican leaders in the Florida Legislature have set aside $4 million for the SNAP contract via an earmark they added to the budget on Sunday, just before finalizing a long-delayed deal on a new state spending plan. The full Legislature is expected to vote on the roughly $115 billion budget on Friday, the final day of a three-week special session in Tallahassee.
The turn toward AI for food assistance is being driven by Republican leaders in Washington, where President Donald Trump and the GOP-controlled Congress are slashing federal support for SNAP, the program formerly known as food stamps.
The federal funding cuts — a combination of eligibility restrictions for individuals and cost-shifting to states — are a key component of the “One Big Beautiful Bill Act,” the sweeping law Trump signed last year that also contains an enormous package of tax breaks primarily benefiting wealthy Americans and large corporations.
For example, the federal government currently pays 100 percent of the cost of grocery benefits provided to families under SNAP. But it will soon begin forcing states to shoulder some of those costs themselves.
How much each state must cover will depend on their payment “error rates” — the percentage of SNAP payments in which a recipient receives more or less than they should. States with payment error rates above 6 percent will have to have to absorb between 5 and 15 percent of the grocery benefit costs.
Only nine states had error rates below 6 percent in 2024, according to the U.S. Department of Agriculture, the federal agency in charge of the SNAP program.
Payment errors are largely the result of unintentional mistakes rather than intentional fraud, the USDA says. And those mistakes typically result from inadequate or unreliable “technology, training and staffing,” rather than improper information from people in need of food assistance.
The federal cuts could fall especially hard on Florida, which had a SNAP payment error rate of just over 15 percent in 2024 — the fourth-highest in the nation. If the state’s error rate remains above 10 percent, Florida could lose nearly $1 billion in support from Washington.
Anti-poverty advocates in Florida have been warning about the looming cuts for months, calling the situation a “four-alarm fire” and urging state legislators and the DeSantis administration to get serious about identifying and addressing the underlying causes of over- and underpayments.
Activists fear that if Florida doesn’t solve the problem proactively, state leaders in Tallahassee may respond to the resulting loss of federal funding simply by making more cuts — either by reducing grocery benefits for everyone enrolled in SNAP or by kicking some people off the program entirely.
More than 1.4 million households across Florida — more than 2.5 million individual Floridians — rely on SNAP to afford groceries.
It’s against that backdrop that Floria lawmakers now want to spend $4 million bringing in a consultant who can incorporate AI into the state’s SNAP program.
The procurement provision is tucked inside the budget for the Department of Children & Families, the state agency that administers SNAP in Florida. It specifically charges the vendor with helping get the state’s error rate below 6 percent — “to a level that eliminates the state share of liability for such errors.”
In addition to using machine learning to review eligibility decision decisions, lawmakers want the vendor to identify root causes of payment errors and recommend changes to avoid future mistakes.
Florida is not the only state going down this path. The state of West Virginia, for instance, recently awarded a nearly $1 million contract to SAS Institute, a data analytics firm, for AI-enabled software intended to help reduce the state’s SNAP error rate. SAS has also lobbied Florida lawmakers on the issue.
But there is an entire constellation of vendors now peddling similar technology in state Capitols across the country. Companies such as Deloitte, Gainwell Technologies and Anthropic are all touting AI-enabled software that they say can help states lower their SNAP error rates and maintain federal funding.
It is, perhaps, a helpful reminder that bureaucratic changes to major public-benefit programs are often, as much as anything else, money-making opportunities for contractors and consultants.
Executives at one such firm — Maximus Inc. — have crowed to Wall Street that they expect double-digit profit growth as a result of Trump’s One Big Beautiful Bill, which, in addition to the assortment of SNAP changes, also imposes so-called “work requirements” on the Medicaid health insurance in many states.
“The legislation creates meaningful opportunities for Maximus in both Medicaid and SNAP,” Maximus President and CEO Bruce Caswell told investment analysts on a company earnings call late last year.
“We’re cautiously optimistic that this increased urgency around SNAP will lead to procurement activity,” Caswell added.
There isn’t a lot of detail in the five-sentence instructions that accompany the proposed contract procurement. So it’s not yet clear whether Florida would use AI solely to flag SNAP applications and eligibility decisions that may warrant further human review — or whether it could let AI make such decisions itself.
The state of Michigan has begun using artificial intelligence to review SNAP applications, according to reporting by the Michigan Advance, though state officials told the publication that staff “remains responsible for all case decisions.”
Spending $4 million on an outside consultant isn’t the only step Florida lawmakers are taking to prepare for the impending SNAP changes.
Another provision in the budget orders the Department of Children & Families to begin making monthly reports to the Governor’s Office and legislative leaders starting Aug. 1 with updates on the ongoing error rate, detailed breakdowns of mistakes, and identification of “primary factors” contributing to errors, among other information.
But Florida lawmakers also punted on some more difficult decisions.
For example, DCF also asked the Legislature for nearly $50 million for upgrades to the aging automated eligibility system it currently uses to process applications for SNAP, Medicaid, and the cash-assistance program known as Temporary Assistance for Needy Families or TANF.
GOP legislative leaders cut funding for the modernization project from the final budget.
Not a good look at all. We are allowing AI to run our lives without any guardrails. These are people, who need food to live. A machine has no thoughts about anything, much less hunger.
The testimony from 20- something DOGE staffers who used AI to cut grant funding for programs that included anything DEI is a clear warning for this proposal. It’s too easy to accept AI results without verifying, and misguided prompts and political animus can do great harm.