10 to the 23 AI logo
Work With 1023AI

Domain Atlas

Public benefits & eligibility

Fraud scoring, eligibility automation, and care allocation — the domain with the largest documented harms, almost all of them ending in courts and commissions.

What AI is doing here

Unemployment fraud adjudication

Predictive

Automated determination of benefits fraud from data matching, with penalties and collections downstream.

Welfare fraud risk scoring

Predictive

Ranking recipients for investigation priority by modeled fraud likelihood.

Eligibility processing automation

Predictive

Automated intake, document processing, and timeliness rules replacing caseworker-managed eligibility.

Care-hours allocation

Predictive

Algorithmic assessment setting home- and community-based care hours for disabled and elderly recipients.

What has gone wrong — and right

Documented deployments, presented as stylized model organizations with full citations.

Michigan MiDAS

Michigan, USA

Automated unemployment-fraud adjudication with no human review — the canonical high-error, zero-correction structure.

Robodebt (Australia)

Australia

Unlawful income-averaging debt assessment at national scale, with the burden of proof reversed onto recipients.

Indiana / IBM eligibility modernization

Indiana, USA

Privatized, automated benefits-eligibility processing that produced mass denials and ended in contract collapse and litigation.

Rotterdam welfare-fraud risk model

Rotterdam, Netherlands

A machine-learning fraud-risk score whose audited bias led a city to suspend it — and a rare look inside a live scoring system.

Arkansas ARChoices / ARIA

Arkansas, USA

Algorithmic allocation of home-care hours that cut care for many recipients and lost repeatedly on due-process grounds.

Who is in the system

  • Agency leadership. Owns procurement, policy, and the authority map; answers for the system publicly.
  • Served people & families. Those the decisions land on. Deliberately outside the PAN dynamics — their outcomes are measured, never simulated.
  • Vendors. Build and update the systems; hold the information asymmetry procurement must govern.
  • Regulators & oversight bodies. Boards, auditors, data-protection officers, inspectorates — external correction capacity.
  • Courts & commissions. The heaviest, slowest actors — who end most of the failures documented in this Atlas.
  • Advocates & community organizations. Surface harms institutions do not see; historically the earliest accurate signal.

Dominant pressures

  • Reviewer bottleneck. One fixed-capacity checking stage sits between AI output and consequence; everything queues behind it.
  • Austerity & recovery incentives. Cost-cutting and overpayment-recovery targets tilt the system toward denial and enforcement errors.
  • Vendor opacity. The deploying institution cannot inspect the model, data, or update pipeline it is accountable for.
  • Compliance over substance. Paper controls (sign-offs, checklists) satisfy audits while the behavior they describe erodes.

Questions leaders should be asking

  1. 1. Which direction of error does the system's design actually minimize — and who chose that?
  2. 2. Can an affected person reconstruct why the decision happened, well enough to contest it?
  3. 3. Is there a pre-authorized way to pause the system, or does stopping require litigation?
  4. 4. Who reviews the automated adverse actions nobody appeals?

For the actions behind these questions, see the Practice Library.