Michigan's automated unemployment fraud detection system MiDAS falsely accused over 40,000 people of fraud with a 93% error rate between 2013-2015, resulting in wrongful seizure of benefits, wages, and tax refunds.
Between October 2013 and August 2015, Michigan's Unemployment Insurance Agency deployed the Michigan Integrated Data Automated System (MiDAS), a $47 million automated system designed to detect unemployment fraud and replace human fraud investigators. The system operated with minimal human oversight and automatically flagged discrepancies between employer and employee reported income as evidence of fraud. MiDAS sent online questionnaires to suspected claimants giving them only 10 days to respond, and failure to respond or unsatisfactory responses automatically triggered fraud determinations. The system then initiated aggressive collection activities including wage garnishment and seizure of tax refunds, imposing the nation's highest penalties of 400% of the alleged overpayment plus interest. A subsequent review found that MiDAS generated approximately 40,000 incorrect fraud determinations out of about 44,000 total cases, representing a 93% error rate for cases decided solely by the automated system. At least 20,000 people had money wrongfully seized, with the state eventually refunding $21 million. Many victims filed for bankruptcy, lost jobs due to background checks, and suffered damaged credit ratings. The system was discontinued for automated fraud determinations in September 2015 after federal pressure and multiple lawsuits, with the state reaching a $20 million settlement in 2022.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that fail to perform reliably or effectively under varying conditions, exposing them to errors and failures that can have significant consequences, especially in critical applications or areas that require moral reasoning.
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed