Convicted by Prediction: Why Pre-Crime Policing Breaks the Law It Serves

June 7, 2026

Picture, for a moment, the file. It does not exist on paper. It exists as a row in a database held on a server somewhere in the National Police Chiefs' Council estate, on a Home Office machine in Marsham Street, or in the back end of a contractor's analytics platform racked in a data centre on the edge of a Reading business park. The row contains a name, an address, a list of associations and a risk score. The man whose name sits in the first column does not know the row is there. He has not been arrested, charged, cautioned or interviewed. He has not been told that an algorithm has assessed his propensity for predatory violence against women and girls and returned a number high enough to place him in the top one thousand most dangerous men in England and Wales. He cannot ask to see the file. He cannot appeal its conclusions. He may, however, find that the police know his car, his routine and his ex-partner's address before he has met the constable on his doorstep. The file precedes him.

This is V1000, the proposal that broke into the British public sphere in January 2026 when the Telegraph reported that Sir Andy Marsh, head of the College of Policing, was advocating the use of predictive analytics to identify the one thousand men deemed most likely to commit predatory offences against women and girls before any such crime had been committed. The scheme modelled itself on the Met's V100 programme, launched in summer 2023, which uses a points-based scoring system to rank the hundred London men assessed each month as posing the greatest risk to women. By autumn 2025 V100 had produced over 200 convictions with sentences totalling more than 676 years. V1000 is the same logic scaled tenfold and pushed nationwide, embedded in a Home Office white paper that Home Secretary Shabana Mahmood unveiled in late January 2026 as the most significant reorganisation of British policing in two centuries. In the same round Mahmood reached for the line about “the eyes of the state” being “on you at all times,” a sentence that invokes Bentham's panopticon and that, as Silkie Carlo of Big Brother Watch has long argued, does not belong in a healthy democracy.

The panopticon line is not the most consequential thing the British state has said about predictive policing in the past eighteen months. The most consequential thing it has done is build systems that go further than V1000 contemplates, and do so largely without telling the public. In April 2025 Statewatch published freedom-of-information documents showing that the Ministry of Justice had been quietly developing a Homicide Prediction Project, since renamed “sharing data to improve risk assessment.” Commissioned under Rishi Sunak's premiership in January 2023, it draws on records held by the Ministry of Justice, the Home Office, Greater Manchester Police and the Metropolitan Police, ingests data on between 100,000 and 500,000 people, and was designed to model who was most likely to commit murder. The contract documents specifically identified mental health, addiction, self-harm, suicide history, vulnerability and disability as variables expected to give the model “significant predictive power.” Sofia Lyall, the Statewatch researcher who led the work, described it as “the latest chilling and dystopian example” of British state crime-prediction, a tool that would “reinforce and magnify the structural discrimination underpinning the criminal legal system.” A previous Ministry of Justice tool, the Offender Assessment System known as OASys, had already been shown to produce less accurate predictions for Black offenders than for white ones.

A government is framing predictive policing, in public, as a solution to a serious category of violent crime. In practice it is constructing infrastructure that does substantially more than the framing acknowledges, with forces whose underlying data has been repeatedly shown by their own regulators to be racially skewed. The question the Telegraph's January 2026 reporting forces is what kind of legal order can accommodate such systems without ceasing to be a legal order at all.

The Inventory of the American Experiment

Across the Atlantic, the Brennan Center for Justice published on 20 November 2025 a report titled The Dangers of Unregulated AI in Policing, authored by Rachel Levinson-Waldman, director of the Center's Liberty and National Security Program, and Ivey Dyson, counsel in that programme. The report is an inventory of the systems police departments across the United States have adopted, in most cases without public debate, legal frameworks governing accuracy, or mechanisms for the surveilled to contest their inclusion. It names the New York City, Los Angeles, Chicago, Boston, Pasco County Sheriff's Office and Washington DC Metropolitan police departments as forces that have deployed AI-driven data-fusion platforms to compile risk profiles and direct enforcement. It documents that 80 to 90 per cent of investigated ShotSpotter gunfire alerts in the cities where the system has been studied have produced no confirmed gun-related offence. It records that at least eight of the ten wrongful arrests known to have been based on facial recognition involved Black individuals. It notes that over 95 per cent of Suspicious Activity Reports forwarded to the FBI between 2010 and 2017 were never investigated, which means the act of generating, ingesting and storing the report, with all its downstream consequences for the person reported, was sufficient injury in itself.

The Brennan Center's argument is not that any single component is faulty. It is that the combination of components, the absence of accuracy standards, the opacity of procurement, and the inability of the surveilled subject to challenge the conclusions drawn about them, together produce a regime the United States constitutional tradition has no vocabulary for. The November 2025 report extends Levinson-Waldman's decade of work on police surveillance to the data-fusion era, where the question is no longer whether a given algorithm predicts crime accurately but whether the assemblage of inputs, scoring, surveillance and consequence functions as an extralegal apparatus that bypasses the protections the rest of criminal procedure was built to enforce.

The American case studies do not require imagination. Chicago ran the Strategic Subject List, colloquially the heat list, from 2012 onwards, assigning everyone it identified a score representing their assessed risk of involvement in gun violence. Robert McDaniel, a Black man then aged twenty-two and living on the South Side, received an unannounced visit from a police commander in late 2013 warning him not to commit further crimes. McDaniel's prior record consisted of a marijuana-possession charge and an illegal-gambling offence. He had attracted attention not for violent conduct but because of where he lived and whom he knew. The visit was sufficient, in his account and in the record assembled by reporters at the Verge, to mark him in his neighbourhood as a police informant. He was shot and wounded shortly afterwards. He was shot at again years later. The heat list was discontinued in early 2020 after a RAND Corporation audit found the early programme had no measurable preventative impact on gun violence and that its principal observable consequence had been a heightened concentration of police contacts on those whose names appeared on it.

In Pasco County, Florida, the Sheriff's Office ran its Intelligence-Led Policing programme, in which a computer system identified people predicted to commit future crimes, including many under eighteen. Deputies were instructed to make frequent “prolific offender checks,” which in practice meant arriving at the door, photographing the household, citing the resident for minor infractions like uncut grass, and returning at intervals. The Institute for Justice filed a federal lawsuit in 2021 on behalf of four residents, including Dalanea Taylor, Tammy Heilman and Robert A. Jones III. In December 2024 the Sheriff's Office settled, paid $105,000, and accepted that the programme had exceeded officers' implied licence to knock on doors, interfering with the plaintiffs' First, Fourth and Fourteenth Amendment rights. It is one of the few US legal proceedings in which a court extracted a clear finding that a predictive policing programme had violated constitutional rights, and only because the office settled rather than risk a precedent-setting trial.

The COMPAS recidivism-risk algorithm, used in pre-trial bail and sentencing across the United States, was the subject of a 2016 ProPublica investigation by Julia Angwin, Jeff Larson, Surya Mattu and Lauren Kirchner that compared COMPAS scores assigned to more than 7,000 people arrested in Broward County, Florida, with their actual subsequent offending. Black defendants were almost twice as likely as white defendants to be incorrectly flagged as high risk while not actually re-offending; even controlling for criminal history, age and gender, they were 77 per cent more likely to be classified as higher risk of future violent offending. Eric Loomis, whose Wisconsin appeal reached the State Supreme Court in 2016, had no meaningful way to inspect the algorithm or challenge his score because it was a trade secret of a private firm and the court accepted that contention. The court upheld the score's use while cautioning that future cases might raise due-process violations if judges did not understand the tool's limits. The caution was not operationalised in any subsequent precedent. The tool remains in use.

What “Pre-Crime” Means When the Statute Catches Up

In February 2026 the USC Dornsife Scribe published an analysis by Jerry Wood, The Pitfalls of Predictive Policing in the Minority Report, that extended the comparison the Telegraph's coverage had invited. The Philip K. Dick story, first published in 1956 and adapted by Steven Spielberg in 2002, imagines a world in which three precognitive humans foresee murders before they occur and the state arrests the would-be perpetrators on the strength of the forecast. The fictional system's conceit is that it works, in the narrow sense that those arrested would, in the absence of intervention, have committed the crimes attributed to them. Real predictive policing systems carry no such guarantee. They are statistical, probabilistic and unverifiable: the prediction's accuracy cannot be tested without permitting the predicted event to occur, and the prediction's effect on subsequent behaviour cannot be cleanly separated from the effect of the police intervention it triggers.

The Dornsife piece reaches back to scholars including Sarah Brayne, whose 2020 ethnography of the LAPD's use of Palantir Gotham, Predict and Surveil, documented how the platform fused arrest records, license-plate reads, field-interview cards, gang databases, foreclosure records, vehicle registrations and noise complaints into a single interface that extended police gaze into every artefact of municipal life. Brayne's central observation is that the platform did not introduce new biases so much as ratify and amplify the biases already encoded in the underlying records, with the additional property that the ratification appeared, to its users, to be objective and authoritative.

Andrew Guthrie Ferguson, whose 2017 book The Rise of Big Data Policing remains the most thorough legal account of the field, has made a parallel argument about the problem algorithmic policing poses for American criminal procedure. The Fourth Amendment protects against unreasonable searches but does not obviously regulate the construction of a database that renders a person more likely to be searched in future. The Fourteenth Amendment's Due Process Clause protects against the deprivation of liberty without due process, but the liberty interest in not being stigmatised by a state-held risk score has, with the partial exception of the Pasco settlement, not been recognised as cognisable. The Equal Protection Clause demands evidence of discriminatory intent, which is rarely demonstrable in an algorithm's designers, while the discriminatory effect of biased training data is attributed by the algorithm's defenders not to the algorithm but to the world it describes. The American constitutional vocabulary was not built for the problem.

The British position is different in detail and, in some respects, more permissive of executive action. The United Kingdom lacks a single written constitution and operates through a combination of common-law principles, the Human Rights Act 1998, the Data Protection Act 2018, the Equality Act 2010, and the supervisory authority of the Information Commissioner's Office. The Gangs Matrix maintained by the Metropolitan Police, on which 79 per cent of those listed as of late 2021 were Black, was the subject of an ICO enforcement notice in November 2018 finding it in serious breach of data-protection legislation, and a 2022 judicial-review settlement in which the Met accepted that the matrix had been operated unlawfully. The settlement created, for the first time, a right for those listed to request access to their inclusion, but did not extend to a substantive right of challenge, and the matrix continued to operate in modified form. Amnesty International UK's Automated Racism report of 20 February 2025 found that at least thirty-three police forces across the UK were operating predictive profiling or risk-prediction systems in “flagrant breach” of national and international human-rights obligations because they were being used to racially profile people and to undermine the presumption of innocence by targeting them before any crime had been committed.

The Regulator Wakes, Slowly

The AICerts coverage of February 2026 captured a moment in which regulators in multiple jurisdictions began to confront, in coordinated rather than fragmentary fashion, the growing evidence that predictive policing systems were not merely imperfect but structurally biased. The European Union's AI Act, whose Article 5 prohibitions came into force on 2 February 2025, includes at Article 5(1)(d) a categorical ban on AI systems that “assess or predict the risk of a natural person committing a criminal offence, solely on the basis of profiling or assessing personality traits and characteristics.” The operative word is “solely,” which European AI lawyers have read as carving out systems that combine profiling with at least one element of “objective and verifiable” evidence linked to criminal activity. The carve-out, narrow on its face, is wide in practice. Almost any predictive system in operation, including any conceivable V1000 successor, can be characterised by its operators as drawing on objective inputs in addition to profiling. The European Data Protection Supervisor and a coalition of civil-society organisations have called for the carve-out to be tightened. The lobbying continues; the systems continue to operate.

In the United States the regulatory landscape is more fragmented. The White House Office of Management and Budget issued in 2024 a memorandum requiring federal agencies to conduct impact assessments for “rights-impacting” AI uses, including in law enforcement. The memorandum does not apply to state and local police departments, which conduct the overwhelming majority of policing. New York City's POST Act requires the NYPD to publish impact and use policies for surveillance technologies; the Brennan Center has argued that the policies published in compliance are so generic and so devoid of operational detail that they impede rather than enable meaningful public oversight. In February 2026 the Department of Homeland Security signed a blanket purchase agreement, reported by AICerts and several other outlets, valued at up to $1 billion for data-fusion software, an order of magnitude that compresses the federal procurement timeline below the speed of any plausible regulatory response.

The pattern is consistent. Departments procure predictive systems on operational rationales emphasising efficiency. They deploy them before the frameworks that govern them are drafted. They publish, at best, impact assessments after deployment. They reform at the margins in response to litigation. They continue, in substance, to use them. The regulatory pace is slower than procurement by years; procurement is slower than the technology by months. The accumulation is of systems whose operation runs ahead of the legal vocabulary needed to discipline them.

The Feedback Loop Is the Architecture

The most consequential observation in the AICerts February 2026 reporting, and in the wider literature it summarises, is that predictive policing systems do not merely inherit historical bias in their training data. They constitute and reinforce that bias as a feature of their operation. The mechanism is well-documented. Place-based systems, of which PredPol was the most widely deployed in the 2010s, assess the likelihood of crime in a given location by reference to the recorded crime in that location. The recorded crime in a location is the product, in significant part, of the police presence in that location. When the algorithm directs additional police to a high-risk location, the additional observation generates additional recorded crime, which feeds back into the model as confirmation that the location is, indeed, high risk. The loop has been demonstrated mathematically by Kristian Lum and William Isaac, whose 2016 paper modelling PredPol on Oakland drug-arrest data showed that the algorithm would concentrate police attention in neighbourhoods where police had previously concentrated, regardless of the underlying distribution of drug use, which independent survey data showed to be roughly uniform across racial groups.

Person-based systems exhibit a parallel pattern. A score, once assigned, attracts police attention. That attention generates contacts, citations, arrests and intelligence reports, all of which feed the next score. The trajectory is not falsifiable from inside the system, because the system has no access to ground truth about what the person would have done absent the intervention. The USC Dornsife analysis of February 2026 framed the issue as one in which the algorithm “does not predict future behaviour so much as amplify past enforcement patterns.” The system reads the history of policing as the history of crime, the demographics of policed neighbourhoods as the demographics of criminality, and the absence of records from less-policed neighbourhoods as the absence of crime there. The output is not a forecast in any scientific sense. It is a re-presentation, in a vocabulary that carries the unearned prestige of mathematics, of the existing pattern of state attention.

The implications for V1000 are direct. The V100 draws on police records of prior incidents, intelligence reports, calls for service, witness statements and patterns of association. Each is shaped by the prior history of policing in the geographies from which they are drawn. The V100's reported success in producing convictions does not establish that the algorithm has identified the men who pose the greatest risk. It establishes that the algorithm has identified men against whom the police have been able to mount successful prosecutions, a related but distinct quantity. The Met has not disclosed false positive rates. It has not disclosed the demographic composition of the ranked cohort. It has not published an equality impact assessment specific to V100. The infrastructure on which V1000 will be built is one in which the most basic accuracy and fairness metrics are unpublished, the inputs are systematically shaped by the prior pattern of British policing, and the consequences of inclusion are, for the subject, materially significant and procedurally unchallengeable.

The Right Not To Be Predicted

What does due process require in the age of pre-crime prediction? The answer is not, despite the Minority Report comparison V1000 has invited, a categorical prohibition on statistical methods in policing. Police forces have always made resource-allocation decisions on the basis of pattern, intelligence and judgement. The question is what procedural protections must surround the use of automated systems that assign individual risk scores with material consequences for the people scored. A defensible regime requires, at minimum, the following.

The first requirement is notice. A person placed on a predictive watch list, assigned an individual risk score, or otherwise subjected to algorithmic risk assessment by a state agency must be told. The principle is foundational to procedural fairness in every developed legal system. It is, in the case of predictive policing, the requirement most uniformly violated. V1000 contemplates no notice. The Homicide Prediction Project contemplates no notice. The Gangs Matrix did not contemplate notice until the 2022 settlement forced a limited right of subject-access. The American systems documented by the Brennan Center contemplate no notice. The absence of notice forecloses every subsequent procedural protection, because the subject cannot challenge a process they do not know is happening.

The second requirement is access. The subject must be entitled to inspect the inputs used to generate the score, the weights assigned to them, and the reasoning by which the score was reached. The trade-secret defence asserted by Northpointe in the Loomis litigation, accepted by the Wisconsin Supreme Court, is incompatible with this requirement, and the Loomis precedent is increasingly viewed as a failure of judicial nerve. Where the algorithm is the product of a private vendor, the answer is not to defer to the vendor's commercial interest but to require, as a condition of public procurement, the disclosure of the algorithm and the underlying data to the subject and counsel.

The third requirement is challenge. The subject must have a substantive right of appeal, before an independent body, with the power to remove the subject from the list if the inputs are inaccurate, the inferences unjustified, or the algorithm itself shown to be discriminatory. The 2022 Gangs Matrix settlement created a right of subject-access without a meaningful right of substantive challenge. The Pasco settlement extracted a commitment to discontinue the programme but did not establish a generalisable right of challenge for similar programmes elsewhere. The EU AI Act creates rights of explanation for individuals affected by high-risk AI systems but excludes the systems used by law-enforcement and migration agencies in ways that render the protections substantially weaker for precisely the populations most subject to algorithmic harm.

The fourth requirement is audit. Police forces and ministries that deploy predictive systems must publish, on a regular cycle, accuracy and fairness metrics broken down by demographic group, and must subject the systems to independent evaluation by bodies with the technical capacity and legal authority to demand the underlying data. The RAND evaluation of Chicago's heat list is the prototype. It is also, fifteen years into the era of person-based predictive policing in the United States, almost the only such evaluation that has been published. The dearth is not coincidence. Audit threatens the operational autonomy of the agencies deploying the systems and the commercial value of the vendors supplying them. It is, for both reasons, systematically resisted. The remedy is statutory mandate.

The fifth requirement is proportionality. A tool that secures convictions of people who have already offended is a tool for prosecution. A tool that prevents offences before they occur is of a different and more constitutionally fraught character. The Met's V100 has, on the public record, secured convictions. It has not been shown to have prevented offences that would otherwise have occurred. Conflating the two is a category error V1000's public advocates have, throughout the white-paper process, declined to address.

The sixth requirement is reversibility. Where a predictive system has affected a person, the harm must be capable of being undone. A wrongful inclusion on a watch list, once acted upon, can produce harms that no subsequent administrative correction can reach. McDaniel's inclusion on the Chicago heat list, the police visit that announced it to his neighbours, and the shootings that followed are not events the eventual discontinuation of the programme could undo.

The Limits of the Architecture

These requirements, even if implemented in full, would not resolve every problem predictive policing presents. They would leave open the more fundamental question of whether some categories of state action are simply incompatible with a free society regardless of the procedures attached. The argument that V1000, the Homicide Prediction Project, the Pasco programme and the Chicago heat list share a common defect that no procedural architecture can repair is the argument civil-liberties organisations on both sides of the Atlantic have been making for the better part of a decade. The defect is the substitution of statistical inference for the substantive legal process by which a state is permitted to deprive a person of liberty. It is categorically incompatible with the presumption of innocence and with the requirement that punishment follow from the proof of an act rather than the prediction of one.

The Brennan Center, the USC Dornsife scholars, Amnesty International UK, Statewatch and Big Brother Watch have all reached the same operational conclusion. The current predictive-policing infrastructure does not meet the requirements of due process under any plausible reading of either constitutional tradition. The systems are deployed without notice, without access, without challenge, without audit, without demonstrated proportionality, and with effects that cannot be made reversible. The result, on the ground, is a regime in which a person can be placed on a list, surveilled, visited, photographed, cited, harassed and, in the worst cases, killed, on the basis of a model whose accuracy they cannot test, whose inputs they cannot inspect, and whose conclusions they cannot contest. This is not the rule of law. It is something else, wearing the rule of law's clothes.

The choice between V1000 and its alternatives is not a choice between safety and rights. It is a choice about which kind of safety, for which population, secured by which means, at the cost of which rights, for which other population. The men whose names will appear on the V1000 list will not be a representative sample of the men in England and Wales who pose a risk to women. They will be a sample whose composition reflects the patterns of British policing's prior attention. The list will, in the aggregate, generate convictions, because lists drawn from the records of police attention have always been able to generate convictions when police attention is renewed. The convictions will be cited as evidence the list works. The men wrongly included will not appear in the statistics. The crimes the list fails to prevent, by directing attention away from offenders whose patterns do not match the algorithm's training distribution, will not appear in the statistics either. The performance of the system will be measured by its consonance with itself.

The women whom V1000 is designed to protect have a separate set of interests. They have an interest in being protected from the men who pose risks to them, which is the interest the scheme's advocates have placed at the centre of the public case. They have, equally, an interest in a criminal-justice system whose treatment of suspects and convicted persons does not so corrode the legitimacy of state power that its eventual response to actual violence is rendered less, rather than more, effective.

The Standard the Moment Requires

A mature legal order would, faced with the V1000 proposal, have set the conditions of its operation in advance. It would have required the publication of the algorithm and its training data, at least to the Information Commissioner and to designated independent reviewers. It would have required an equality impact assessment, conducted before deployment and refreshed annually. It would have required notice to those placed on the list, with a substantive right of appeal to an independent tribunal. It would have required statutory limits on the actions police could take on the basis of inclusion, with particular protections for inputs derived from third-party data such as health, school or social-services records. It would have required regular external audit of accuracy, bias and operational outcomes. It would have required, before national rollout, evidence of demonstrable preventative effect in the form of a controlled comparison with non-algorithmic alternatives. It would have required, as a backstop, a sunset clause that withdrew the legal authority for the programme if the evidence of effectiveness did not materialise.

None of these conditions, on the public record as of late May 2026, have been set. The white paper announcing V1000 contains no published algorithm, no equality impact assessment, no notice mechanism, no appeal right, no statutory limit on consequential police action, no external audit framework, no controlled pilot evaluation, no sunset clause. The Telegraph's January 2026 reporting captured the moment at which a substantial expansion of British algorithmic policing was announced in advance of the procedural protections that would have rendered it constitutional in either the European or the American sense. The Brennan Center's November 2025 inventory, the USC Dornsife analysis of February 2026 and the AICerts coverage of the same month establish that the British announcement is the latest instance of a pattern, not an outlier.

The constitutional question is not whether the algorithm is accurate. It is whether the people whose lives it rearranges have any meaningful say in the rearrangement. They do not. Until they do, the systems being built in Britain and the United States, and increasingly in the European Union notwithstanding the AI Act's nominal prohibitions, are not predictive instruments in any rigorous sense. They are administrative instruments for the redistribution of state attention, dressed in the prestige of computation, that operate beyond the reach of the procedural protections the rest of the criminal-justice system, at least nominally, requires. The Minority Report comparison, which V1000's public advocates have treated as a rhetorical excess from civil-liberties campaigners, captures something the public advocates have not addressed. In the Dick story, the system worked. In the world the Telegraph described in January 2026, the system does not need to work to do harm. It needs only to be believed. The belief is the architecture, and the architecture is being poured.

What due process requires, then, is the recovery of a principle older than the technology that threatens it. The principle is that the state may act against a person on the basis of what they have done, after a process in which they can know the case against them, see the evidence, and answer it. The principle is not consistent with secret lists, secret scores, secret models and secret consequences. It does not bend because the technology has become sophisticated enough to make the bending operationally efficient. The men on the V1000 list, the people in the Brennan Center's American inventory, the residents whose lives the Pasco programme reorganised, the Black Londoners whose names the Gangs Matrix held, and the future subjects of systems yet to be procured all have the same basic claim. They have the right to know, the right to see, the right to challenge, and the right, before the state visits their door, to a process. The current generation of predictive systems treats that claim as administrative friction. The treatment is the failure. The recovery of the claim is the work.

References

Telegraph reporting on UK predictive policing plans, January 2026, as relayed via secondary coverage including GB News, “Police chiefs could trial 'Minority Report policing' to identify and catch criminals before they strike,” January 2026. https://www.gbnews.com/news/police-chiefs-trial-minority-report-policing-identify-catch-criminals-before
TechInformed. “Government bets on AI to predict and prevent crime.” 2026. https://techinformed.com/uk-government-bets-on-ai-to-predict-crime/
UK Government. From local to national: a new model for policing. White Paper, January 2026. https://assets.publishing.service.gov.uk/media/69779267276692606c013862/260125_White_Paper.pdf
Rachel Levinson-Waldman and Ivey Dyson. The Dangers of Unregulated AI in Policing. Brennan Center for Justice, 20 November 2025. https://www.brennancenter.org/our-work/research-reports/dangers-unregulated-ai-policing
Jerry Wood. “The Pitfalls of Predictive Policing in the Minority Report.” USC Dornsife Scribe, 27 February 2026. https://dornsife.usc.edu/scribe/2026/02/27/the-pitfalls-of-predictive-policing-in-the-minority-report/
AI CERTs News. “Regulators Confront Predictive Policing Bias.” February 2026. https://www.aicerts.ai/news/regulators-confront-predictive-policing-bias/
Amnesty International UK. Automated Racism: How police data and algorithms code discrimination into policing. February 2025. https://www.amnesty.org.uk/files/2025-02/Automated%20Racism%20Report%20-%20Amnesty%20International%20UK%20-%202025.pdf
Statewatch. “UK: Ministry of Justice secretly developing 'murder prediction' system.” April 2025. https://www.statewatch.org/news/2025/april/uk-ministry-of-justice-secretly-developing-murder-prediction-system/
Metropolitan Police. “Groundbreaking technology boosts Met's fight against violence towards women and girls.” https://news.met.police.uk/news/groundbreaking-technology-boosts-mets-fight-against-violence-towards-women-and-girls-498976
Metropolitan Police. “V100 Violence Against Women and Girls Programme data and impact assessments.” Disclosure October 2025. https://www.met.police.uk/foi-ai/metropolitan-police/disclosure-2025/october-2025/data-impact-assessments-mps-v100-violence-against-women-girls-programme/
Information Commissioner's Office. Enforcement notice issued to the Metropolitan Police Service, Data Protection Act 1998. November 2018. https://www.met.police.uk/SysSiteAssets/media/downloads/force-content/met/about-us/gangs-violence-matrix/ico-enforcement-notice.pdf
Youth Justice Legal Centre. “Metropolitan Police to overhaul 'racist' Gangs Matrix after landmark legal challenge.” https://yjlc.uk/resources/legal-updates/metropolitan-police-overhaul-racist-gangs-matrix-after-landmark-legal
Julia Angwin, Jeff Larson, Surya Mattu, Lauren Kirchner. “Machine Bias.” ProPublica, 23 May 2016. https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing
State v. Loomis, 881 N.W.2d 749 (Wis. 2016).
Jessica Saunders, Priscillia Hunt and John S. Hollywood. “Predictions Put Into Practice: A Quasi-Experimental Evaluation of Chicago's Predictive Policing Pilot.” RAND, 2016. https://www.rand.org/pubs/external_publications/EP67204.html
Techdirt. “How Predictive Policing Got A Chicago Man Shot Twice.” 3 June 2021. https://www.techdirt.com/2021/06/03/how-predictive-policing-got-chicago-man-shot-twice/
Institute for Justice. “Case Closed: Pasco Sheriff Admits 'Predictive Policing' Program Violated Constitution.” Press release. https://ij.org/press-release/case-closed-pasco-sheriff-admits-predictive-policing-program-violated-constitution/
Sarah Brayne. Predict and Surveil: Data, Discretion, and the Future of Policing. Oxford University Press, 2020. https://global.oup.com/academic/product/predict-and-surveil-9780190684099
Andrew Guthrie Ferguson. The Rise of Big Data Policing: Surveillance, Race, and the Future of Law Enforcement. NYU Press, 2017. https://nyupress.org/9781479892822/the-rise-of-big-data-policing/
Kristian Lum and William Isaac. “To predict and serve?” Significance, October 2016. https://rss.onlinelibrary.wiley.com/doi/10.1111/j.1740-9713.2016.00960.x
European Union. Regulation (EU) 2024/1689 (the AI Act), Article 5 prohibitions, applicable from 2 February 2025. https://artificialintelligenceact.eu/article/5/
Future of Privacy Forum. “Red Lines under the EU AI Act: Unpacking the Prohibition of Individual Risk Assessment for the Prediction of Criminal Offences.” https://fpf.org/blog/red-lines-under-the-eu-ai-act-unpacking-the-prohibition-of-individual-risk-assessment-for-the-prediction-of-criminal-offences/
Big Brother Watch. Director profile of Silkie Carlo and campaigns on facial recognition and predictive systems. https://bigbrotherwatch.org.uk/

Tim Green UK-based Systems Theorist & Independent Technology Writer

Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.

His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.

ORCID: 0009-0002-0156-9795 Email: tim@smarterarticles.co.uk

Listen to the free weekly SmarterArticles Podcast

Discuss...