British Police Built a Sprawling Crime-Prediction Machine. Some Results Couldn’t Be Trusted

British police created a sprawling system of machine-learning models that assigned risk scores to hundreds of thousands of people based on sensitive data including police records, housing status, mental health information, and other personal details, without most residents knowing about it. An investigation found that at least two of these risk-scoring models were quietly abandoned after officials determined they could no longer be trusted, and independent reviewers warned the systems lacked transparency and could undermine public trust. The situation matters because the UK is now moving toward wider adoption of predictive analytics across the criminal justice system, with leadership from the former chief of the police force that built this experiment. People affected by these systems often did not know whether they had been scored by an algorithm, what their score was, or how it might affect their interactions with authorities.