Michigan's Unemployment Benefits Algorithm MiDAS Issued False Fraud Claims to Thousands of People

Michigan’s MiDAS system falsely accused over 34,000 people of unemployment fraud from 2013 to 2015, which reportedly caused financial ruin for many. The automated system was designed to cut costs, but it adjudicated fraud cases without human oversight. That led to an 85% error rate. Victims faced wage garnishments, some lost homes, and some faced bankruptcy. Despite early warnings, Michigan’s UIA defended MiDAS until lawsuits and federal pressure forced reforms. Legislators have been seeking compensation for those wrongfully accused.

Risk classification

• Primary risk domain: 5 Human-Computer Interaction
• Primary risk subdomain: 5.2 Loss of human agency and autonomy

The state delegated key decision-making authority regarding fraud determinations and financial seizures entirely to the MiDAS algorithm, removing human review and leaving victims with no meaningful immediate recourse or appeal.

Additional risk subdomains

• 7.3 Lack of capability or robustness: The MiDAS system failed to perform reliably, operating with a catastrophic 93 percent error rate due to flawed mathematical assumptions and an inability to verify data context.

Causal factors

• Entity: AI
• Intent: Unintentional
• Timing: Post-deployment

The severe financial and personal harms were caused by the automated decisions of the MiDAS system, which produced unexpected and highly erroneous outcomes while pursuing the goal of fraud detection.

EU AI Act risk tier

• Risk tier: 2 High Risk

High Risk: The system is an AI used by a public authority to evaluate eligibility for public assistance benefits and services, which has significant implications for safety, fundamental rights, and public interests.

AI system and alleged parties

• AI system: Michigan Integrated Data Automated System (MiDAS)
• AI purpose: Cheating Detection; Application Evaluation
• Behaviour type: Autonomous
• Alleged developer: Fast Enterprises, CSG Government Solutions
• Alleged deployer: Michigan Unemployment Insurance Agency
• Alleged harmed parties: unemployed Michigan residents falsely accused of fraud, Michigan residents who faced bankruptcy or foreclosure due to MiDAS

Harm severity

Highest direct severity in any category: Substantial. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Minor
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Substantial, indirect Substantial
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, indirect Negligible
• Differential treatment: direct Substantial, indirect Negligible
• Civil rights: direct Substantial, indirect Negligible
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Substantial
• Epistemic: direct Negligible, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Physical

Reported: The reports describe indirect physical harm resulting from the financial distress caused by the incident.

Directly caused: N/A

Indirectly caused: At least one claimant was forced to forgo essential diabetes treatments due to bankruptcy and lack of funds, and others experienced homelessness after home foreclosures.

Inferred additional harm: It is highly likely that other individuals among the 40,000 falsely accused suffered physical health declines or engaged in self-harm due to extreme stress, though these are not quantified.

Financial

Reported: The reports describe extensive financial losses, including a 20 million dollar class-action settlement and over 11,000 bankruptcies.

Directly caused: The state wrongfully seized millions of dollars in tax refunds and garnished wages from tens of thousands of claimants, with an average financial loss per occurrence of approximately 20 million to 30 million dollars in aggregate seized funds.

Indirectly caused: Claimants suffered indirect losses from legal fees, high interest rates due to ruined credit, and bankruptcy filing costs. Over 11,000 families filed for bankruptcy.

Inferred additional harm: The total financial devastation, including lost wages from job opportunities denied due to fraud flags on background checks, likely exceeds the 20 million dollar settlement by tens of millions.

Differential treatment

Reported: The reports describe systemic differential treatment targeting unemployment claimants.

Directly caused: The automated system targeted unemployment claimants, a vulnerable socioeconomic group, subjecting them to automated, error-prone fraud determinations without the due process afforded to other citizens.

Indirectly caused: N/A

Inferred additional harm: Socioeconomic inequality was exacerbated as the poorest and jobless residents bore the brunt of the system's failures.

Civil rights

Reported: The reports explicitly describe widespread violations of constitutional and civil rights.

Directly caused: The system directly violated the constitutional due process rights of approximately 40,000 citizens by seizing their property (tax refunds and wages) without notice, human review, or a meaningful opportunity to be heard.

Indirectly caused: N/A

Inferred additional harm: N/A

Psychological

Reported: The reports explicitly describe severe psychological and mental-health trauma caused by the false accusations.

Directly caused: N/A

Indirectly caused: Thousands of claimants suffered severe emotional distress, anxiety, and trauma. The mental health toll was so severe that a university legal clinic added a suicide hotline to its referral page to assist desperate clients.

Inferred additional harm: Widespread, unquantified psychological trauma and anxiety likely affected a significant portion of the 40,000 falsely accused individuals and their families.

People affected

• Occurrences reported: 1
• People reportedly harmed: 40000
• People reportedly exposed: 62784

Potential causes

Management

• Prioritizing Cost Over Accuracy: Focused on reducing staff and cutting budgets instead of system safety.
• Ignoring Warning Signs: Management defended system because of the surplus revenue it generated.
• Contractor Compliance Focus: Developers followed orders blindly without questioning system validity.

Technology

• Robo-Adjudication Algorithm: System made final fraud decisions without any human oversight or review.
• Flawed Earnings Calculation: Divided quarterly income by 13 weeks to falsely assume steady weekly work.
• Inability to Assess Intent: Algorithm could not determine if discrepancies were due to simple mistakes.

Data Inputs

• One-Sided Data Preference: Programmed to automatically assume the employer data was always correct.
• Missing or Corrupt Data: System based some of its fraud findings on missing or corrupt database records.
• Complex Government Databases: Engineers misunderstood complex data structures and collections processes.

Human Factors

• Engineering Hubris: Engineers overestimated capabilities and ignored complex local workflows.
• Automation Bias of Overseers: Human overseers were hesitant to question or override automated results.
• Confusing Portal Notifications: Claimants missed pings sent to inactive online accounts instead of mail.

Process and Methods

• Elimination of Human Staff: Laid off one-third of the agency workforce, removing human oversight.
• Flawed Questionnaire Design: Confusing multiple-choice questions led innocent people to self-incriminate.
• Inadequate Appeals Window: Strict 30-day window to appeal was finalized before claimants knew of charges.

Regulatory Environment

• Highest-in-Nation Penalties: State law mandated 400 percent penalties, amplifying the financial harm.
• Lack of Independent Vetting: No independent interdisciplinary group vetted technology before purchase.
• Deficient Due Process Protections: Absence of legal requirements for pre-deprivation human review.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide extensive, consistent, and detailed coverage of the MiDAS system, its deployment, the legal proceedings, the number of affected individuals, the error rate, and the specific harms caused. Multiple reputable sources corroborate the core facts.
• Ambiguities identified: The exact technical details of the algorithm's code are not fully public, but its operational logic is well-documented through court filings.
• Alternative interpretations: None that are plausible; the role of the automated system in causing the harm is clear and legally established.

The MiDAS incident represents a major failure of an autonomous state-level decision-making system that violated the civil rights of 40,000 citizens. While it caused severe local economic and societal distress, its direct national security implications remain negligible to minor.

• Overall national security impact: Minor
• Response level: Moderate
• Scope: Single nation
• Primary target: United States
• Alleged perpetrator: Michigan Unemployment Insurance Agency

Threat characteristics

• Imminence: Long-term. The specific incident is resolved, representing a historical precedent and long-term strategic lesson regarding automated government decision-making.
• Autonomy: Full autonomy. The system made fraud determinations, sent notifications, and triggered financial penalties autonomously without human intervention or oversight.
• Novelty: Evolved capability. Represents an early, significant advancement in the deployment of fully automated, unchecked algorithmic decision-making by government entities.

Impact by dimension

• Physical security: Negligible. The incident involved a state-level administrative software system with no impact on physical systems, critical infrastructure, or kinetic military capabilities.
• Information security: Negligible. No intelligence compromise, espionage, or state-sponsored information warfare operations were associated with this domestic administrative failure.
• Sovereignty: Minor. Automated decision-making compromised state-level government functions and violated constitutional due process, but did not threaten national sovereignty or federal operations.
• Economic security: Negligible. While causing severe financial distress to individuals, the incident had negligible impact on national economic systems, strategic industries, or technological security.
• Societal stability: Minor. The system systematically violated the civil and due process rights of approximately 40,000 citizens, causing severe localized societal distress, but did not trigger widespread civil unrest.