UK police built a crime-prediction machine — and some results couldn’t be trusted
an investigation revealed avon and somerset police used 23+ ml models to score citizens for risk, but two were scrapped for unreliable results, highlighting a transparency crisis in predictive policing.
avon and somerset police built a sprawling predictive policing program featuring 23+ machine learning models that scored nearly 500,000 people—using data on mental health, housing, and school meals—to predict offenses like burglary and domestic abuse. a wired investigation found that officials quietly scrapped at least two of these risk-scoring models after determining the results could no longer be trusted, while government inspectors warned that the lack of transparency undermines accountability. the disclosure of these 'risk scores' to citizens has been inconsistent, sparking legal challenges from community activists like john pegram who suspect they are on these watchlists without consent. this case signals a growing crisis in law enforcement tech, where the deployment of ai often outpaces validation mechanisms and legal frameworks.