THREAT ASSESSMENT: Institutional Drift Toward Algorithmic Control in U.S. Welfare Systems
Past implementations of algorithmic governance in public assistance have consistently shifted toward enforcement mechanisms, with reversal requiring extraordinary external intervention—typically taking seven to twelve years to correct, as seen in Michigan and other jurisdictions.
Bottom Line Up Front: AI systems in U.S. welfare administration are structurally biased toward enforcement and surveillance over support, with institutional mechanisms making this control-oriented drift routine and reversal exceptionally difficult.
Threat Identification: The threat is the systemic convergence of AI-enabled governance in welfare programs toward control functions—such as fraud detection and risk screening—despite stated goals of support, due to institutional incentives and cost allocation designs that favor measurable cost savings over social assistance outcomes.
Probability Assessment: High likelihood of continued drift toward control in existing and new systems; reversal to support orientation is rare and requires extraordinary external intervention. This trend has been observed across multiple jurisdictions and program types since at least the early 2010s (e.g., MiDAS case) and is accelerating with increased adoption of predictive analytics [Dedyaev, 2026].
Impact Analysis: The consequences include wrongful denials of benefits, erosion of public trust, disproportionate harm to marginalized populations, and entrenchment of punitive welfare models. In the Michigan MiDAS case, over 40,000 individuals were falsely accused of fraud, leading to financial ruin and psychological trauma; the system’s reactivation required only administrative action, while its correction took nine years and a $20 million settlement [Dedyaev, 2026]. Similar risks exist in child welfare (Allegheny Family Screening Tool) and homelessness prevention programs.
Recommended Actions: (1) Redesign cost-allocation frameworks so that the burden of algorithmic error falls on administrators, not beneficiaries; (2) Mandate judicial or legislative review for any algorithmic system that can terminate or restrict entitlements; (3) Establish independent oversight bodies with audit and suspension authority; (4) Prioritize algorithmic transparency and due process in welfare tech procurement.
Confidence Matrix:
- Threat Identification: High confidence (supported by six process-traced cases)
- Probability Assessment: High confidence (historical patterns and institutional logic are well-documented)
- Impact Analysis: High confidence (empirical outcomes from MiDAS, AFST, and others)
- Recommended Actions: Medium-High confidence (inferred from institutional analysis and policy leverage points)
Citation: Dedyaev, M. (2026). Hybrid Algorithmic Governance in U.S. Welfare Administration: State- and County-Level AI as a Case of Support-Control Convergence. arXiv:XXXX.XXXXX [cs.CY].
Published July 7, 2026