Medicaid requires states to pay when disabled or elderly people enter nursing facilities because they need a significant level of care. As an alternative, all states offer home- and community-based services so that people can stay at home with support from paid caregivers to get out of bed, bathe, use the restroom, prepare meals, eat, and go to medical appointments. These home services help disabled and elderly people stay independent and on average cost less than nursing facilities. 

AI systems decide who qualifies for Medicaid, how much in-home care elderly and disabled people receive, and whether Medicaid covers a particular treatment or service. States and their vendors make different design choices about which assessment questions get fed into the algorithm and how they are scored. For this reason, an assessment that would cause one state’s AI tool to decide someone’s care needs are at a level that qualifies them for home and community-based services would result in a denial if the same assessment were fed into another state’s algorithm. Though the specifics of the assessments and algorithms vary, these AI systems inevitably harm recipients. They often result in denials, termination, or reduction of hours of care.