As of April, 37 million people were receiving SNAP benefits, down from a monthly average of 42 million the year before, according to preliminary Agriculture Department data reported by NPR [1]. More than 4 million people are estimated to have lost the benefit between last July and April. The Center on Budget and Policy Priorities puts the national decline over that window at 11 percent [1].
Those three figures do not stack on top of each other, and the reason is worth walking through before anyone quotes them. The 42 million is an average across the prior year's months. The 37 million is a single point, April 2026. Subtracting one from the other gives 5 million, not the reported "more than 4 million," because that subtraction compares an average to a point and answers a different question than the loss estimate answers. NPR's piece does not print a July 2025 participation figure anywhere, which means the loss estimate cannot be reproduced from the numbers on the page.
Run the percentages and only one arrangement holds all three. A 5 million drop from a 42 million base is 11.9 percent. A 4 million drop from that same base is 9.5 percent. Neither is 11 percent. CBPP's 11 percent, applied to April's 37 million, implies a July 2025 starting point near 41.6 million and a loss near 4.6 million, which is the only reading on which "more than 4 million" and 11 percent are both true at the same time. That reconstruction is ours, not a published figure, and the thing to check it against is the July 2025 month in USDA's own series when the department posts it. The direction of all of this is not in question. The precision is.
The rules doing the cutting took effect under the One Big Beautiful Bill Act. Adults now have to document at least 80 hours of work or volunteering a month to keep benefits, and the requirement reaches groups that were previously exempt: veterans, people who are homeless, young adults aging out of foster care, parents whose youngest child is between 14 and 17, and people between 55 and 64 [1]. The Congressional Budget Office projected the work rule would cut average monthly participation by 2.4 million over 2025 through 2034 [1]. A separate provision stripped federal food aid eligibility from refugees, people seeking asylum, and victims of domestic abuse or trafficking [1].
An 80-hour rule is not self-executing. Somebody has to prove the hours, and somebody at a state agency has to receive the proof, read it and act on it. Katie Bergh, a senior policy analyst at CBPP, described what that looks like from the household side: "People are calling and calling, and they can't get through to anyone. Or they're being asked for more and more and more documentation of every aspect of their lives, and maybe they don't have a way to document everything" [1]. The distinction she draws is between need and process. "What that's telling us is that this is not happening because fewer people need help affording groceries," she said. "It's the result of these policy changes" [1].
Arizona is where the change has run furthest. SNAP enrollment there is about half what it was a year ago, more than 400,000 fewer participants [1]. For the first time, according to the Arizona Food Bank Network, more Arizonans visit a food bank in a given month than are enrolled in food stamps [1]. Natalie Jayroe, chief executive of the Community Food Bank of Southern Arizona, described the state's position without much cushioning: "We think of ourselves as the canary in the coal mine. We are showing the rest of the country a really scary scenario" [1]. Texas has lost roughly half a million participants, and NPR lists Louisiana, Florida and Oklahoma among the steepest declines [1].
CBPP reviewed data from 19 states that broke out child enrollment separately. In those states alone, more than 1 million children have lost food benefits since last July [1]. That is a partial count covering fewer than half the states, so the national child figure is higher than 1 million by a margin nobody has published.
The provisions with the largest fiscal consequences have not landed. Starting in October, the federal government's share of state SNAP administrative costs falls from 50 percent to 25 percent, leaving states to carry 75 percent of operational expenses [1]. Those expenses are the staff who answer the phones Bergh's families cannot get through on and who process the documentation those families are being asked for. In October 2027, states begin paying a share of benefit costs themselves if their payment error rate is at or above 6 percent, and CBPP estimates almost half of states could each owe $100 million or more under that penalty [1].
The error rate is not a fraud measure. It counts overpayments and underpayments both, and Bergh characterized it as reflecting "unintentional mistakes by state eligibility workers and participating families," cases where "someone made a typo or a state worker misapplied a policy or a family didn't understand what information they needed to report and when" [1]. A state that underpays a household feeds the same statistic that will start costing it money in October 2027. Underpaying is not the failure mode a penalty is supposed to deter.
State agencies are already pricing that in. In a survey by the Urban Institute and the American Public Human Services Association, to which 39 states responded, 29 percent said they may consider further narrowing eligibility for food assistance and 11 percent said they may need to withdraw from or pause the program altogether if the new costs become too burdensome [1]. Lexie Kuznick, APHSA's director of policy and government relations, called it "really an existential crisis in the future of SNAP" [1].
The administration reads the same numbers as evidence the program was carrying people it should not have been. The Agriculture Department has said improper payments totaled $10 billion last year [1]. Agriculture Secretary Brooke Rollins told Fox Business: "A lot of it is people taking the program that shouldn't have been, and then a lot of it is just a better economy" [1]. The White House called SNAP "bloated" and said it was failing its mission to serve as "temporary help for those who encounter tough times" [1].
That account is testable, and the test is not in this data. If a stronger labor market were moving people off SNAP, exits would concentrate among people who found work and would run fastest where employment is rising fastest. The preliminary release as NPR describes it does not carry that breakdown, and neither does the reporting around it. What the reporting does carry is a state where food bank traffic now exceeds SNAP enrollment, which runs the other way from what a better-economy explanation predicts.
The money does not stop at the household either. The National Grocers Association estimates the decline in shoppers receiving food aid will reduce grocery store sales by nearly $88 billion nationwide through 2034 [1]. A SNAP benefit is spent at a checkout counter, which is where a canceled case file turns into a missing revenue line.
Starsky Wilson, president of the Children's Defense Fund, put the pace at the center of his objection: "We're upset about how quickly this has happened. There are some supports that are still staged to go away later this year" [1]. He warned of "an even greater sense of desperation among children and their families as we come to the end of this year" [1]. On the question of what catches a household that falls out, his answer took one sentence: "There's no replacement for SNAP if a state gets rid of it" [1].
