{"id":61029,"date":"2026-06-26T06:34:44","date_gmt":"2026-06-26T10:34:44","guid":{"rendered":"https:\/\/overcentral.com\/en\/?p=61029"},"modified":"2026-06-26T06:34:44","modified_gmt":"2026-06-26T10:34:44","slug":"uk-police-ai-flagging-errors","status":"publish","type":"post","link":"https:\/\/overcentral.com\/en\/uk-police-ai-flagging-errors\/","title":{"rendered":"UK Police Crime Prediction Model Flags 90% of People Incorrectly"},"content":{"rendered":"

An independent audit of predictive policing models used by Avon and Somerset Police<\/a> in the UK<\/a> has revealed that the vast majority of individuals flagged as high-risk by the systems were incorrectly identified, raising serious questions about the deployment of artificial intelligence in law enforcement. The data, obtained through public records requests and analyzed by AI auditing firm Eticas<\/a>, shows that one model used to predict burglary operated with a precision rating below 10 percent for more than three years, meaning fewer than one in ten people flagged as a likely offender would actually commit a crime.<\/p>\n

Predictive Models Showed Critically Low Accuracy Over Years of Operation<\/h2>\n

The audit examined performance data for 13 risk models used by Avon and Somerset Police between 2017 and 2024. These models were designed to predict a range of outcomes, including missing persons cases, antisocial behavior, and the likelihood of individuals committing or falling victim to crime. According to Eticas, most of the models produced low precision scores, indicating that a high proportion of flagged individuals were incorrectly identified as risks. The findings describe the performance patterns as “not typical of well-governed models in operational use,” with metrics shifting sharply over time.<\/p>\n

A spokesperson for Avon and Somerset Police told WIRED that the force chose not to deploy some of the models it developed, including the one related to burglaries. When asked why years of audit and performance data existed for models that were never used, the spokesperson explained that the audit process was automated and drew from a static file that was not deleted after the decision was made not to deploy the model. The force declined requests for interviews about its data science program and did not fully respond to a detailed list of questions.<\/p>\n

Bias Monitoring Was Fundamentally Inadequate<\/h2>\n

In response to public records requests, Avon and Somerset Police provided a screenshot of a “bias check app” that compared average risk scores for white individuals and people of color, concluding there was “no significant difference between the two.” The Eticas review found this approach deeply insufficient. “Simply including ethnicity as a monitoring variable is not equivalent to testing whether the model produces discriminatory outcomes,” the audit stated, describing the absence of granular testing by ethnicity, gender, and socioeconomic status as “a significant omission.”<\/p>\n

The force’s own ethics committee does not appear to have discussed predictive analytics after 2017, according to records request disclosures. Although the force states on its website that each product and project is reviewed by a dedicated ethics group, the spokesperson confirmed that “there has so far been no meeting held,” because “no model has been produced for which potential ethical issues have been identified.”<\/p>\n

What Does This Mean for Policing and Public Trust?<\/h2>\n

The findings highlight a fundamental challenge at the intersection of data science and law enforcement: models that flag large numbers of false positives risk undermining both operational effectiveness and public trust. If officers are presented with risk scores that are wrong nine times out of ten, the tool ceases to inform judgment and instead introduces noise \u2014 or worse, bias \u2014 into critical decisions about surveillance, intervention, and resource allocation.<\/p>\n

Davies, a former official involved in the force’s data science work, noted that predictive analytics still requires significant refinement. “When we were trying to do it, we were trying to do it for the right reasons, in the right way, but we didn’t have the capacity that it probably needed,” he said. He also warned that there is a risk that staff see a computer-generated prediction and then “don’t use their own judgment.”<\/p>\n

Despite these concerns, predictive analytics continues to be used in the region. Bristol City Council still employs a risk-scoring model to assess the likelihood of a child falling out of education, employment, or training. Avon and Somerset Police’s most recent audit data, provided in July 2024, indicates that the model used by the Offender Management App correctly predicts just one in three people who actually offend, while one in four people flagged as likely offenders do not.<\/p>\n

How Affected Individuals and Communities Should Respond<\/h2>\n

For anyone who may be subject to predictive policing assessments \u2014 whether through contact with social services, offender management, or community policing programs \u2014 the most practical step is to understand your rights under UK data protection law. You have the right to request access to personal data held about you, including any risk scores or algorithmic assessments, through a Subject Access Request (SAR). If you believe a model has produced a discriminatory or inaccurate outcome, you can<\/a> raise a complaint with the force’s data protection officer and escalate the matter to the Information Commissioner’s Office<\/a> (ICO). Privacy advocates also recommend that individuals monitor how risk assessments are used in decisions that affect housing, benefits, or child welfare, and challenge decisions where the rationale is unclear or the underlying data is unreliable.<\/p>\n","protected":false},"excerpt":{"rendered":"

An independent audit of predictive policing models used by Avon and Somerset Police in the UK has revealed that the vast majority of individuals flagged as high-risk by the systems were incorrectly identified, raising serious questions about the deployment of artificial intelligence in law enforcement. The data, obtained through public records requests and analyzed by […]<\/p>\n","protected":false},"author":7,"featured_media":84360,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"fifu_image_url":"https:\/\/cards.overcentral.com\/cards\/en\/61029.png","fifu_image_alt":"UK Police Crime Prediction Model Flags 90% of People Incorrectly","footnotes":""},"categories":[349],"tags":[],"class_list":["post-61029","post","type-post","status-publish","format-standard","has-post-thumbnail","category-articles"],"fifu_image_url":"https:\/\/cards.overcentral.com\/cards\/en\/61029.png","fifu_image_alt":"UK Police Crime Prediction Model Flags 90% of People Incorrectly","_links":{"self":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts\/61029","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/users\/7"}],"replies":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/comments?post=61029"}],"version-history":[{"count":0,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts\/61029\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/media\/84360"}],"wp:attachment":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/media?parent=61029"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/categories?post=61029"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/tags?post=61029"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}