Minnesota stands up to Trump: it demands to see the algorithm before accepting the $2 billion Medicaid cut

By StateScoop · October 8, 2026.

Minnesota's Department of Human Services is bracing for a serious financial blow: the Centers for Medicare and Medicaid Services (CMS, the federal agency that oversees these public health programmes) has threatened to withhold $2 billion a year in Medicaid funds from the state over alleged fraud, and has already frozen $550 million over recent quarters, including $259 million blocked in February 2026 when the Trump administration announced the crackdown. Interim Medicaid commissioner John Connolly sent a letter last week —reviewed by StateScoop— to CMS Administrator Mehmet Oz demanding something very specific: that the federal government show its work.

According to the letter itself, CMS has identified 870 Minnesota providers as suspected of fraud or anomalous billing practices using "machine learning and artificial intelligence capable of analysing Medicaid claims". The problem, Minnesota says, is that it does not know on what basis: it is asking CMS to spell out the methodology, the criteria that turn a provider into a "high risk" one and the specific technology used, because without that evidence it cannot decide whether to suspend payments or open its own investigation. Connolly stresses that the state is willing to halt payments when the evidence supports it —something he says it does routinely— but is still waiting for an answer from the agency on the path to a resolution, despite having already completed its corrective action plan.

The backdrop is not fictional: real fraud exists and has already reached the courts. The Justice Department operation known as the Minnesota Health Care Fraud Takedown ended in February with criminal charges against 15 people, among them daycare owners and Medicaid providers, and last week two men pleaded guilty to diverting more than $500,000 from the state's Housing Stabilization Services programme through false claims. In response, Governor Tim Walz signed a bipartisan law in June creating a governance board to modernise human services systems, funding technology upgrades and opening grants for counties to improve their own systems. In March, Trump had also created a federal anti-fraud task force focused on the benefits system as a whole —housing, food, health care, cash assistance— of which this Medicaid case is just one piece.

What separates this episode from a simple budget tug-of-war is the nature of the evidence. CMS is not presenting court rulings or case files: it is presenting the output of an algorithmic analytics system and asking the state to act on that basis, without opening up the model. It is the same pattern we have been flagging in other areas of AI applied to the public sector: the forum in which a claim is made determines its reliability. A press release boasting of "billions in detected fraud" is not subject to the same scrutiny as a methodology that would have to survive an audit or a courtroom. Minnesota is asking, rightly, for the second to exist before accepting the first. State officials themselves put it bluntly: CMS "blurs the line between actual criminal fraud and administrative actions", and they expect it to do so again in its next announcement about the programme.

This matters beyond Minnesota. The case is a dress rehearsal for how AI is going to mediate, increasingly, decisions that move billions of dollars in public money and affect the health coverage of millions of people. If the pattern repeats without transparency being demanded —the algorithm flags, the money is frozen, the provider or the state has to prove its innocence without seeing the evidence— the burden of responsibility is reversed before any real oversight framework exists. The risk is not abstract: a report from the Center on Budget and Policy Priorities, a Washington-based think tank, warns that the new Medicaid work requirements set out in last year's budget reconciliation law could cause millions of beneficiaries to lose coverage not for breaking the rules, but for being unable to navigate outdated or overloaded administrative systems. When the underlying infrastructure is fragile, automating the detection of "irregularities" on top of it multiplies false positives, not just hits.

In the short term, this is going to be an uncomfortable bumpy patch: states litigating with Washington over access to the models that decide on their money, legitimate providers caught in "high risk" lists generated by systems nobody outside the agency can audit, and Medicaid patients as collateral victims of a technical dispute. But Minnesota's own response —a modernisation law, a governance board, investment in its own systems— points to where this has to go: not towards less AI in public administration, but towards an AI legally required to explain its criteria. That is the version of abundance we want to defend here: public systems that use AI to manage limited resources better, freeing up real money for health care instead of for litigation, but only if auditability is built alongside algorithmic capability, not years after the damage is done.