The news website Axios has revealed that US state governments are increasingly moving to integrate artificial intelligence technologies into the management of social safety-net programmes, with the aim of reducing workloads and streamlining bureaucratic procedures for government employees. The website noted that this technological shift is being driven by federal legislation pushing states into the digital age, with some having already launched AI-powered chatbots to answer queries from Medicaid beneficiaries, while Florida lawmakers included an AI-based eligibility verification system for the Supplemental Nutrition Assistance Programme (SNAP) in the 2027 budget, and New Hampshire is collaborating with Google Gemini to simplify unemployment benefit application processes.
With stricter work requirements and more frequent periodic reassessments for Medicaid and SNAP applicants set to take effect next year, states are seeking to hand these time-consuming tasks to machines rather than human workers. However, experts and civil rights advocates are warning that adequate safeguards around these technologies are absent, noting that AI systems can hallucinate and commit serious errors that may deprive citizens of essential life-sustaining benefits.
Health policy analysts stress that a minor discrepancy in statistical data — such as slight differences in reported income between assistance applications — could automatically result in the exclusion of genuinely eligible individuals from benefit programmes.
Past legal cases highlight the scale of risks associated with these algorithms. Arkansas faced a successful lawsuit after using an automated system to determine home care entitlements, which resulted in arbitrary and harmful reductions to the eligible hours of thousands of residents. During the litigation, it emerged that the authorities had made no projections about the scale of the cuts, and had no staff capable of explaining the mechanical workings of the algorithm, which was subsequently found to operate in illogical ways that did not meet patients' actual care needs.
Oversight challenges are compounded by AI developers' refusal to disclose how their software operates, citing proprietary data and commercial privacy protections — leaving affected individuals without a safety net for several months even when their eligibility is legally reinstated.
In the face of these challenges, technology and public policy experts are emphasising the need to subject these programmes to continuous testing and monitoring in safe, limited pilot environments before rolling them out at the state level. They stress that while pursuing greater administrative efficiency is important, passing erroneous decisions against the most vulnerable groups is unacceptable, and that officials must either scrutinise these technologies rigorously or continue to rely on traditional means of managing social support case files.