BBC essay says AI policing can feed biased records back as predictions

BBC News on October 25, 2026 published an edited essay arguing that police use of AI can deepen discrimination against minority groups. The essay, by Adrian Dredge, is the entry that won the 2026 Young Science Writer Award from the Association of British Science Writers. It is a reported walk through existing research, plus interviews, and it does not present a new dataset. The page’s published time is 12:18 UTC on October 25. This account is not legal advice.

What Brayne reported from Los Angeles

The essay leans on Sarah Brayne, a sociologist at the University of Texas at Austin, and her fieldwork with the Los Angeles Police Department. In “Big Data Surveillance: The Case of Policing,” published in the American Sociological Review in 2017, Brayne writes that historical crime data fed to predictive tools are systematically biased: crimes in public are more visible, some people are less likely to report, and police focus attention and resources on Black communities at a disproportionately high rate relative to drug use and crime rates. She writes that once those records are entered, “the predictions appear impartial.” She then describes predictive models as performative, “creating a feedback loop in which they not only predict events such as crime or police contact, but also contribute to their future occurrence.”

The BBC essay uses those findings to describe hotspot maps that darken where officers already spend time, because more presence produces more recorded crime. That mechanism is Brayne’s argument in the 2017 paper. The BBC piece applies it to current AI tools as an essay, and it does not name a new deployment or a new audit.

What the NYU authors counted

The essay also cites researchers at New York University on “dirty data,” records shaped by corrupt, biased, or unlawful policing. The underlying article is “Dirty Data, Bad Predictions,” by Rashida Richardson, Jason M. Schultz, and Kate Crawford, in the NYU Law Review in 2019. The authors say they identified thirteen jurisdictions where public records showed an overlap between predictive-policing development or use and government investigations, consent decrees, or other documentation of corrupt, racially biased, or otherwise illegal practices. They write that the analysis “revealed nine jurisdictions” where police data from periods of unlawful and biased practices “was available to train or otherwise inform predictive policing systems.” They say that overlap presented at least some risk that the systems could be influenced by those practices, or in some cases perpetuate them. They highlight Chicago, New Orleans, and Maricopa County as case studies, and they say vendor assurances do not, in their view, show that the systems adequately separate this data.

The BBC’s sentence is shorter: in nine of the thirteen US areas, data from periods of unlawful or biased policing was available to train the systems. The count of nine and thirteen is in the paper. The paper’s claim is about availability and risk, which is the wording above.

Two researchers, as the BBC quotes them

The essay quotes Carolyn McKay, an associate professor at the University of Sydney Law School, who told the BBC: “The public functions of state policing ought to be subject to human oversight and scrutiny.” The BBC presents her view as a call to check the data forces use and to remove bias, and it says she remains concerned that place-based systems can still track the people who live in heavily policed areas.

Dimitrios Kollias, an associate professor at Queen Mary University of London, is quoted pushing the other way on one point. He told the BBC that the idea of a system merely copying old maps “has its limits,” because “modern AI systems do not simply retrieve or copy their training examples.” He also told the BBC that more autonomous systems need behavior that “remains safe, reliable and under meaningful human control,” and that bias from dirty data is “an important concern” while “this does not mean that bias is inevitable.” Those sentences are his comments to BBC News, as the essay prints them.

Why it matters (analysis)

The news on October 25 is the BBC’s decision to publish the essay, and the interviews attached to it. The studies it depends on are from 2017 and 2019. A reader who wants the counts should use the NYU paper’s sentence about data being available to train or inform systems, and Brayne’s sentence about a feedback loop, rather than treat the essay as a fresh census of police AI. McKay’s oversight line is her recommendation. Kollias’s comments are his reading of what newer systems can do. Both are labeled as such in the BBC piece. For other policy files, see China’s employment opinion and the AI policy topic.

This report is not legal, financial, or health advice. See the disclaimer.