AI in policing may worsen discrimination

Researchers from British and Australian universities warn that AI used by the police could worsen discrimination, according to an essay by Adrian Dredge published on October 25, 2026, by BBC News. This article, awarded by the Association of British Science Writers, focuses not on temporary software flaws but on the foundational data these tools rely on.

AI inherits historical biases

The AI deployed in policing would inherit decades of unevenly distributed police checks, without being able to detect this imbalance in the archives. These archives record the facts observed by patrols, not the total number of offenses committed.

Biases in data collection

Sociologist Sarah Brayne, who observed the police in Los Angeles, noted two converging biases. Violations committed in public spaces leave more written traces, while the attention of officers has been disproportionately focused on Black neighborhoods. The model therefore learns a geography of crime that first reflects that of patrols.

Dirty data in AI systems

Lawyers from New York University have given a name to this problem, dirty data, literally meaning unclean or biased data. In thirteen American jurisdictions equipped with AI, this issue has been identified.