Domain Atlas / Public benefits & eligibility
Michigan MiDAS
Explore this deployment in the PAN Lab ↗
In the PAN Lab, the readouts of each model organization drawn from this case carry a shaded evidence band whose width follows the least-established class among the modeling inputs the readings rest on.
The least-established modeling input behind the derived readings of the “MiDAS-class automated determination system” model is assumed: “This models the automated-determination pattern documented in the Michigan MiDAS case file — not a reconstruction of the actual system.” Evidence base: 3 assumed · 2 published baseline.
The least-established modeling input behind the derived readings of the “MiDAS-class review network after the settlements” model is assumed: “This models the SAME deployment as the MiDAS-class automated determination network, drawn at its second documented instant: after the settlements that ended the no-review mode. It is a stylized model of the documented post-settlement structure, not a reconstruction of the actual system.” Evidence base: 6 assumed · 3 published baseline.
Michigan's MiDAS system auto-adjudicated unemployment-insurance fraud with an extremely high error rate among automated determinations, wrongly accusing tens of thousands of people; litigation and court action forced review and compensation.[4]
What happened
Michigan's Integrated Data Automated System (MiDAS) determined unemployment-insurance fraud algorithmically, issuing accusations, penalties, and collections in many cases without human review of the determination. Reporting, the state Attorney General's office, and incident documentation describe an extremely high error rate among the auto-adjudicated fraud findings and tens of thousands of people wrongly accused, with wage garnishments and tax-refund seizures downstream. Litigation (the Bauserman line of cases) and court action ultimately forced review and compensation.
The sociotechnical reading
In system-map terms, MiDAS removed the operator network almost entirely: model output flowed to enforcement records and collections with no correction stage. With the error-to-record pathway wide open and the correction dial near zero, errors were self-sustaining by construction — no model improvement could have compensated for the missing human loop. The eventual controls came from actors outside the deploying agency: courts and litigation, years after deployment. The case is the strongest documented argument for why "who can correct, and how fast" matters more than headline accuracy. The Lab draws this deployment at two documented instants: the no-review mode, and the post-settlement structure in which the court-ordered review and the correction of the record are the live pathways.
The concepts used in this reading are defined in the Field Guide; the governance responses live in the Practice Library. The model organization for this case can be stress-tested in the PAN Lab.