AIBRAI:Fair AI, for all.

Roughly 26,000 families were falsely flagged as fraudsters; many were forced to repay tens of thousands of euros and lost their homes, and parliamentary investigations found that hundreds of families were broken apart, with children placed in care as a consequence of wrongful fraud designations. The Dutch parliament concluded that the tax authority's algorithmic risk model used nationality and dual-citizenship as risk indicators, in violation of anti-discrimination law.

The AMS Algorithm classified jobseekers into A/B/C groups that determined access to training budgets. Women, people with care responsibilities, migrants and people with disabilities received automatic deductions, channelling them into the lowest-support group. The Austrian DPA suspended the system in 2020.

Researchers led by Ziad Obermeyer showed that at any given risk score, Black patients were significantly sicker than white patients. Remedying the bias would raise the share of Black patients flagged for extra care from 17.7% to 46.5%. Similar tools were estimated to affect roughly 200 million people per year.

Critics labeled the approach "hurt first, fix later": claimants were investigated and benefits paused before bias was assessed. DWP's own internal documentation acknowledged statistically significant disparities, prompting renewed parliamentary scrutiny.

Lead plaintiff Mary Louis, a Black voucher holder with 16 years of on-time rent, was rejected by SafeRent's score. The algorithm overweighted credit scores and ignored the value of housing vouchers, disproportionately harming Black and Hispanic applicants. The settlement rolled back parts of the screening product.

Udbetaling Danmark and ATP run a sprawling network of fraud-prediction models that flag claimants for invasive investigations. Amnesty concluded the system amounts to mass surveillance and risks violating the rights to social security, privacy and non-discrimination under EU law.

The findings sparked global debate over algorithmic fairness in criminal justice, influenced later research on disparate error rates, and became a reference case for regulators drafting rules on high-risk AI in public-sector decision-making.

Williams was held nearly 30 hours despite an obvious alibi. The case became a landmark for civil-liberties groups: Detroit later settled, agreed to restrict its use of facial recognition as the sole basis for arrest, and required corroborating evidence before charges.