The Global Evolution of Predictive and Preemptive Law Enforcement

Full report

On this page

Executive Summary

This report delivers an exhaustive, multi-regional analysis regarding the adoption of technologies and institutional practices engineered to anticipate crime, violence, terrorism, public disorder, and future security risks. Synthesizing original legislation, procurement records, court decisions, and independent investigations, the analysis yields fifteen principal findings. The first principal finding establishes as a verified fact that global law enforcement has undergone a paradigm shift, transitioning from place-based crime forecasting toward person-based algorithmic risk assessment \[Confidence: High\]. While early systems focused on geographic hotspots, modern architectures target specific individuals with intensive, predictive scrutiny based on their historical data. Secondly, it is a reasoned inference that the core driver of modern preemptive policing is not algorithmic innovation, but rather the unprecedented mass integration of previously siloed government and commercial databases \[Confidence: High\]. The capability to fuse disparate data sets has fundamentally altered the state's surveillance capacity. Thirdly, independent findings reveal a pervasive, strategic rebranding effort across the commercial surveillance industry. As explicit claims of predictive efficacy have faced intense public pushback and academic debunking, disgraced predictive systems are routinely repackaged under euphemisms such as "decision support," "precision policing," or "resource routing"1 \[Confidence: High\]. Fourth, it is a verified fact that democratic and authoritarian governments increasingly deploy identical technical architectures—frequently procured from the same corporate oligopoly—albeit adapted for different institutional and political purposes2 \[Confidence: High\]. Fifth, the analysis identifies the integration of generative artificial intelligence and foundation models as a verified, revolutionary frontier in intelligence work. Tools utilizing Large Language Models (LLMs) are accelerating the transition toward highly automated, multimodal surveillance and investigative prioritization5 \[Confidence: Medium\]. Sixth, it is a verified fact that the scope of data utilized in predictive policing has expanded exponentially; systems no longer rely solely on criminal records, but actively ingest civil, financial, educational, health, social media, and immigration data to generate threat scores1 \[Confidence: High\]. Seventh, independent findings demonstrate that private-sector vendors frequently invoke commercial confidentiality and trade secret protections to block independent audits of their algorithmic systems, an opacity that severely undermines public accountability and legal due process10 \[Confidence: High\]. Eighth, as a direct consequence of litigation and public backlash, it is a verified fact that explicit numerical risk scores (e.g., the 1-500 scale previously used in Chicago) are being rapidly replaced by opaque "analyst recommendations" or "priority tiers" to minimize legal liability11 \[Confidence: High\]. Ninth, independent findings repeatedly and unequivocally demonstrate that actuarial and machine-learning risk assessments reproduce, automate, and amplify historical policing biases, particularly concerning race, class, and neighborhood demographics14 \[Confidence: High\]. Tenth, while legislative bodies are attempting to construct oversight mechanisms—most notably the European Union’s AI Act—it is a verified fact that these frameworks often feature expansive carve-outs for national security and law enforcement, severely limiting their practical restriction on state surveillance17 \[Confidence: High\]. Eleventh, in highly militarized or conflict-oriented jurisdictions, the collection of predictive data has been gamified. For instance, independent findings verify that military personnel have been incentivized to collect non-consensual biometric data to map civilian populations, accelerating the deployment of automated control systems19 \[Confidence: High\]. Twelfth, it is a reasoned inference that preemptive systems operate on an inverted evidentiary standard, shifting law enforcement interventions away from a standard of reasonable suspicion of past crimes toward the statistical probabilities of future, uncertain conduct \[Confidence: High\]. Thirteenth, it is a verified fact that global law enforcement architectures have become deeply dependent on a concentrated oligopoly of cloud infrastructure providers, creating novel vulnerabilities regarding data sovereignty and national security \[Confidence: Medium\]. Fourteenth, independent findings confirm that despite tens of millions of dollars in continuous investment, the predictive validity of many flagship systems remains statistically negligible, frequently failing to generate actionable intelligence that reduces violent crime12 \[Confidence: High\]. Finally, the ultimate reasoned inference of this report is that worldwide law enforcement is not moving toward a genuine, clairvoyant "pre-crime" capability; rather, it is utilizing the aesthetic of prediction to legitimize radically more intensive, data-driven surveillance and automated prioritization of marginalized populations \[Confidence: High\].

Functional Taxonomy and Definitional Distinctions

To ensure analytical rigor, it is a verified necessity to define and distinguish the specific functional categories of preemptive law enforcement technologies. The conflation of these terms often obscures the epistemological differences between various surveillance systems. Predictive policing constitutes the application of analytical techniques—particularly quantitative data analysis—to identify targets for police intervention and prevent crime or solve past crimes by making statistical predictions about future events22. This is fundamentally distinct from proactive policing, which is a traditional, analog law enforcement philosophy focused on preventing crime through physical police presence and community engagement, though this philosophy has been increasingly digitized. Intelligence-led policing serves as a broader management model that uses data analysis and criminal intelligence to objectively direct police resources toward priority targets and systemic threats, acting as the bureaucratic precursor to fully automated systems. Algorithmic threat assessment refers to the specific use of machine learning or complex statistical models to evaluate the danger posed by a specific individual, event, or object24. When applied to individuals, it manifests as person-based risk scoring, which generates quantitative risk scores or tiers for specific people, ostensibly reflecting their mathematical likelihood of future criminal involvement as either perpetrators or victims1. Conversely, place-based crime forecasting uses historical crime data—primarily the time and location of past incidents—to predict where specific crimes, such as burglaries or vehicle thefts, are most likely to geographically cluster in the future22. Behavioral threat assessment departs from historical crime data to analyze psychological, linguistic, or kinetic behaviors, attempting to determine if an individual poses an imminent risk of violence or terrorism. Watchlisting is the institutional practice of maintaining databases of individuals subjected to heightened scrutiny, restrictions, or surveillance based on intelligence criteria, often used in border screening or counterterrorism. Network and association analysis involves the computational mapping of social, familial, or criminal connections to identify key actors within a network, frequently utilizing co-arrest data to predict future vectors of violence27. Recidivism prediction utilizes actuarial or algorithmic tools within the criminal justice and penal systems to estimate the likelihood that a convicted offender will commit a new crime, a metric that heavily influences bail, sentencing, and parole decisions28. Traveler and border-risk assessment entails the automated screening of passenger manifests, visa applications, and travel histories against intelligence databases to flag individuals prior to or upon arrival at sovereign borders30. Automated surveillance is the continuous, algorithmic monitoring of public or private spaces using sensors, cameras, and data-scraping, operating without the requirement of active human observation10. Within automated surveillance, biometric identification represents the automated recognition of individuals based on biological or behavioral characteristics, most notably through live facial recognition, iris scanning, and gait analysis19. Social-media monitoring is the automated scraping, natural language processing, and analysis of public or semi-public digital communications to map networks, detect civilian sentiment, or identify threats. Anomaly detection employs machine learning systems trained on a baseline of "normal" behavior to flag deviations—such as unusual financial transactions or erratic movement patterns in crowded spaces—for immediate human investigation32. Finally, AI-assisted investigative prioritization represents the use of algorithms to sift through thousands of historical reports and suggest linkages between past crimes, aiding human detectives in identifying serial offenders rather than predicting new ones34.

Global Historical Chronology

The global evolution of data-driven law enforcement can be classified into eight distinct historical and technological phases. Phase 1: Early Statistical Crime Mapping (1990s). The digitization of municipal police records permitted the creation of primitive geographic information systems (GIS). Management frameworks like CompStat in New York revolutionized policing by mapping crime visually to enforce accountability. However, this phase relied entirely on retrospective data, mapping where crimes had already occurred rather than forecasting future events23. Phase 2: Computer-Assisted Intelligence Analysis (Early 2000s). In the aftermath of the September 11 attacks, the creation of intelligence fusion centers demanded databases capable of rudimentary link analysis. Analysts began using software to manually draw connections between disparate actors, telephone numbers, and financial records, moving beyond localized crime mapping into federal and international counterterrorism. Phase 3: Actuarial Risk Assessment (Mid-2000s). Jurisdictions increasingly relied on standardized, point-based surveys to calculate recidivism risk. Tools such as COMPAS digitized these rubrics, introducing proprietary algorithms to the judicial process to assess whether defendants should be granted bail or paroled28. This normalized the concept of using historical data to penalize individuals based on statistical probabilities. Phase 4: Large-Scale Data Integration (Late 2000s \- Early 2010s). The emergence of advanced intelligence platforms enabled the "single pane of glass" view. Software platforms could pull highly siloed records—telecommunications data, motor vehicle registries, criminal warrants, and property records—into interconnected data ontologies. This allowed analysts to search vast data lakes instantly, breaking down the bureaucratic walls that previously separated local, state, and federal intelligence6. Phase 5: Machine-Learning Prediction (2010s). This era marked the shift from descriptive analytics to ambitious predictive models. Place-based systems (utilizing mathematics originally designed to predict earthquake aftershocks) and person-based systems attempted to forecast the future using complex regression and random forest models. Law enforcement agencies openly branded these initiatives as "predictive policing," promising to deploy officers to locations before crimes occurred1. Phase 6: Real-Time Surveillance Analytics (Late 2010s \- 2020s). Predictive algorithms were coupled with live sensor networks. CCTV systems were upgraded with computer vision, integrating live automated facial recognition, automated license plate readers (ALPR), and behavioral anomaly detection. Surveillance transitioned from a forensic tool used after a crime to a proactive filter analyzing crowds in real-time31. Phase 7: Generative-AI-Supported Intelligence Work (2023 \- Present). The deployment of Large Language Models (LLMs) and foundation models began to synthesize intelligence work. Systems emerged that could automatically generate police reports from body-worn camera audio, while advanced intelligence platforms provided natural-language conversational interfaces for highly classified data lakes, vastly accelerating the speed at which intelligence is operationalized5. Phase 8: Multimodal Systems (Emerging). The current and future phase represents the convergence of all previous stages. Real-time video, biometrics, financial history, social networks, and generative AI are integrated into continuous, predictive threat-management pipelines that operate with a high degree of automation. These multimodal systems fuse text, visual, and spatial data simultaneously to generate real-time threat assessments.

Epistemological Distinction: Known Facts vs. Uncertain Conduct

To rigorously assess the validity of preemptive policing, an independent finding emphasizes the critical epistemological distinction between technologies that identify known facts and those that attempt to predict uncertain future conduct \[Confidence: High\]. Systems designed to identify known facts—such as automated license plate readers, basic biometric matching against a watchlist, or historical network analysis (e.g., the NYPD's Patternizr34)—process empirical reality. They match an existing query against an existing database. While these systems can suffer from poor data quality, algorithmic bias, or high false-positive rates, their epistemological foundation is fundamentally sound: they search for evidence of an event that has actually occurred or an individual who definitively exists. Conversely, systems designed to predict uncertain conduct cross a severe epistemological boundary21. Person-based risk algorithms and place-based forecasting models ingest historical data—which is inherently reflective of past police deployment patterns, systemic inequalities, and reporting biases—and utilize it to calculate the mathematical probability of a human action that has not yet materialized. It is a reasoned inference that the conflation of these two categories by law enforcement vendors is a deliberate marketing strategy. By branding speculative predictive systems as "AI-driven" or "data science," vendors grant the aura of empirical, objective certainty to what is, at its core, actuarial guesswork regarding human free will38.

Database of 30 Global Preemptive Law Enforcement Programs

The following matrices document 30 well-verified examples of predictive and preemptive law enforcement programs across more than 20 countries. The data presented represents verified facts, official claims, and independent findings derived from public procurement records, court judgments, legislative actions, and investigative reports.

North America

Europe

Latin America and the Caribbean

Middle East and Africa

Asia and Oceania

World-Region Comparison and Authoritarian Convergence

A global analysis reveals a verified convergence in the technological capabilities of democratic and authoritarian states, albeit constrained by divergent institutional realities and legal frameworks \[Confidence: High\]. In North America and Europe, the deployment of preemptive systems is characterized by robust civil society pushback, intense litigation, and the emergence of nascent regulatory frameworks. Landmark legal battles, such as Bridges v South Wales Police in the UK and the SyRI case in the Netherlands, demonstrate that democratic judiciaries are willing to strike down algorithms that violate privacy or equality laws31. Consequently, the overarching trend in the West is legislative catch-up, highlighted by the European Union’s AI Act17. Because explicit "predictive policing" invites lawsuits, Western vendors rebrand their tools as "resource management" to circumvent stigma1. Conversely, Latin America and Africa are experiencing rapid, uncritical adoption of "Smart City" frameworks. These initiatives are heavily subsidized by foreign telecommunications and surveillance vendors—most notably Huawei in Mauritius and Dahua in South America9. In these regions, data protection laws are frequently bypassed or simply ignored in the name of modernization, public safety, and national security, resulting in opaque public-private surveillance partnerships8. The Middle East and Asia represent the vanguard of unfettered predictive surveillance. From Israel's gamified biometric tracking of Palestinians (Blue Wolf) to China's IJOP in Xinjiang, the technology is fully weaponized for demographic control and warfare20. A critical independent finding is that democratic and authoritarian governments frequently purchase technology from the exact same vendors. Palantir, for example, powers deep-state intelligence in Western democracies (HessenDATA) while Chinese firms (Hikvision, Dahua) supply both European markets and Central Asian autocracies (Kazakhstan's Sergek)2. The underlying technology does not dictate the governance model; rather, the governance model dictates the ethical and legal limits placed upon the technology.

Trend Analysis: The Past 5, 10, and 20 Years

The 20-Year Trend (2006-2026): From Silos to Lakes. Twenty years ago, police intelligence consisted of decentralized, siloed databases. The macro-trend of this era was the relentless pursuit of data integration. The institutional goal was to connect isolated gang databases, motor vehicle records, and warrant registries into a single searchable entity. Intelligence during this period was entirely retrospective; the technology was used to investigate crimes that had already happened. The 10-Year Trend (2016-2026): The Rise and Fall of the "Pre-Crime" Algorithm. The last decade witnessed the zenith of explicit predictive policing. Law enforcement agencies heavily marketed "black box" machine learning algorithms as silver bullets for violent crime. This era saw the peak deployment of place-based systems like PredPol and person-based systems like Chicago's SSL. However, it was also defined by a fierce backlash. Civil rights litigation and academic audits exposed that these algorithms merely laundered historical biases—such as racist arrest disparities—into mathematical mandates, providing a veneer of scientific objectivity to discriminatory policing14. The 5-Year Trend (2021-2026): Multimodal Surveillance and Generative AI. Over the past five years, the field has pivoted away from static risk scores toward continuous, multimodal surveillance. The integration of live CCTV, ALPRs, facial recognition, and social media scraping allows for real-time anomaly detection10. Concurrently, the explosion of Large Language Models (LLMs) has led to systems like Palantir AIP, which allow officers to query massive, classified data ontologies using natural language. This completely automates the analytical process, accelerating the speed at which raw data is transformed into actionable, preemptive interventions5.

Analysis of Program Termination, Rebranding, and Continuation

It is a verified fact that many of the highest-profile predictive policing programs of the 2010s were formally terminated. Chicago's SSL was decommissioned in 2019 after the RAND Corporation proved it ineffective and the Inspector General found it wildly unreliable12. Durham’s HART was scrapped in 2022 due to the massive cost of maintaining ethical compliance and the realization that it created a geographic "postcode lottery" for justice24. However, it is a reasoned inference that program "termination" rarely results in the abandonment of the underlying methodology \[Confidence: High\]. Instead, the surveillance industry relies on strategic rebranding. A premier case study is PredPol. Following intense academic and journalistic exposure demonstrating that PredPol directed police to disproportionately harass minority neighborhoods with an accuracy rate of less than 1%, the company rebranded as Geolitica21. In 2023, the core technology was absorbed by SoundThinking (formerly ShotSpotter) and rebranded again as ResourceRouter21. The fundamental function—using algorithms to direct patrol routes—remains unchanged, but the inflammatory term "predictive policing" has been scrubbed in favor of "community policing foot beat efforts"1. Furthermore, explicit numerical risk scores are being systematically replaced. Because explicitly labeling a citizen as a "Score 450 Risk" invites immediate due process lawsuits and Freedom of Information Act requests, vendors have pivoted. Modern systems, such as those utilizing foundation models, now generate opaque "analyst recommendations" or "priority alerts"6. The algorithm still makes the assessment, but it is cloaked in the language of human-in-the-loop decision support to evade legal liability.

Ten Emerging Trends Likely to Shape the Next Decade

Based on current technological and legislative trajectories, the following ten trends will shape the next decade of preemptive law enforcement \[Confidence: High\]: First, generative AI will become the primary interface for intelligence databases; "AI agents" will synthesize decades of intelligence in seconds, acting as automated analysts. Second, systems like Axon Draft One will normalize AI-authored police reports, raising profound legal questions regarding the admissibility of machine-generated testimony in criminal trials7. Third, global markets will face a "Brussels Effect" as the EU AI Act forces multinational vendors to either build bifurcated models or adopt EU transparency standards globally to maintain market access18. Fourth, to bypass these new civil privacy regulations, law enforcement will increasingly classify domestic predictive systems under "national security" or "counter-terrorism" exemptions18. Fifth, under intense legal pressure, courts will increasingly demand algorithmic explainability, effectively ending the era of proprietary, un-auditable "black box" risk scoring in democratic courts. Sixth, law enforcement will increasingly seek backdoors or data-sharing agreements with domestic Internet of Things (IoT) devices—such as smart doorbells and appliances—to build neighborhood-level predictive models. Seventh, the incentivized, mass collection of biometric data via mobile apps by rank-and-file soldiers and police, as seen in Israel's Blue Wolf, will be exported to other conflict zones and border regions19. Eighth, the reliance of global police forces on a cloud oligopoly (AWS, Google Cloud, Microsoft Azure) will create massive, centralized targets for state-sponsored cyber warfare. Ninth, predictive policing will heavily integrate Central Bank Digital Currencies (CBDCs) and digital payment histories to generate real-time behavioral threat assessments. Tenth, following the precedent set by the 2024 Paris Olympics, continuous algorithmic anomaly detection on public CCTV will become standard, legalized urban infrastructure10.

Ten Areas Where Public Claims Exceed Available Evidence

Vendor and government claims regarding predictive policing frequently outpace empirical reality. The following ten areas highlight where public claims dramatically exceed verified evidence \[Confidence: High\]: First, official claims that predictive software causes drops in crime are routinely debunked; independent findings show crime drops are often correlative with broader macroeconomic trends, not algorithmic interventions15. Second, vendors claim algorithms remove human prejudice, yet empirical evidence proves they codify and automate historical biases present in the training data16. Third, regarding the accuracy of geospatial predictions, companies like PredPol claimed high accuracy, while independent audits revealed success rates of less than 1% in predicting actual incidents21. Fourth, systems are sold as "decision support" keeping a "human in the loop," but psychological automation bias means human operators almost invariably defer to the machine's recommendation without critical thought48. Fifth, claims that data is securely "de-identified" are frequently overstated; modern link analysis can easily re-identify individuals through metadata. Sixth, there is virtually no verified evidence that placing an individual on a "Strategic Subject List" actually deters them from committing future crimes; it merely increases their likelihood of arrest for petty infractions69. Seventh, vendors claim a commitment to transparency, yet routinely invoke trade secrets to prevent public audits of their source code when challenged in court10. Eighth, claims of 99% accuracy rates for Live Facial Recognition (LFR) deliberately ignore staggering false-positive rates when the technology is deployed in uncontrolled, real-world crowds31. Ninth, police departments claim AI saves money, however, the ongoing licensing, cloud storage, and maintenance costs often vastly exceed traditional analytical budgets24. Tenth, systems claiming to detect "suspicious behavior" (e.g., loitering, erratic movement) generate overwhelming noise and false positives, primarily targeting the unhoused or neurodivergent10.

Final Assessment: "Pre-Crime" vs. Data-Driven Surveillance

The central question remains: are global law enforcement agencies moving toward a genuine, clairvoyant "pre-crime" capability? Based on the rigorous synthesis of technical, legal, and sociological evidence, the answer is definitively negative \[Confidence: High\]. It is a reasoned inference that the core mathematics of predictive policing have fundamentally failed to achieve clairvoyance. Human behavior, embedded in vastly complex socioeconomic realities, is not a physical phenomenon akin to weather patterns or earthquake aftershocks1. Algorithms cannot predict the future; they can only regurgitate the statistical patterns of the past. However, the failure to predict the future has not slowed the adoption of the technology. Instead, worldwide law enforcement is moving toward radically intensive, data-driven surveillance and the automated prioritization of citizens. The true power of systems like Palantir, Cortex, and Skynet does not lie in telling police exactly what will happen tomorrow. Their unprecedented power lies in their ability to instantly aggregate every recorded aspect of a citizen's life—tax records, border crossings, social media posts, ALPR hits, and biometrics—and present it in a unified, mathematically ranked dashboard6. By rebranding these vast surveillance architectures as "predictive," "AI-driven," or "precision policing," state apparatuses justify the continuous, unchecked expansion of their data-gathering mandates. The danger of the next decade is not that a computer will accurately predict a citizen committing a crime. The profound danger is that the computer will inaccurately label a citizen a threat based on flawed, biased data, and the bureaucratic machinery of the state—shielded by the opacity of commercial trade secrets and national security exemptions—will act upon that label with unquestioned, automated authority.

Works cited

1. BIG DATA-DRIVEN PREDICTIVE POLICING: GLOBAL PRACTICES AND LESSONS FOR ASEAN COUNTRIES \- AJEE Journal, https://ajee-journal.com/upload/attaches/att\_1773154758.pdf

2. Kazakhstan Is Building A Surveillance State. Will China Be Its Model? \- Radio Free Europe, https://www.rferl.org/a/china-surveillance-hikvision-dahua-huawei-kazakhstan-ai-central-asia/33544093.html

3. UNIVERSITÀ DEGLI STUDI DI PADOVA, https://thesis.unipd.it/retrieve/b904890b-fbf7-4dc7-8c6d-b193b18a930c/The%20Algorithmic%20Borders\_Tesi\_Simona\_Santino\_30.06.26pdfA\_.pdf

4. What Is Palantir? The Company Behind Government AI Tools | Built In, https://builtin.com/articles/what-is-palantir

5. AIP for Defense \- Palantir, https://www.palantir.com/platforms/aip/defense/

6. Inside Palantir AIP: How the World's Most Controversial AI Platform Actually Works, https://towardsai.com/p/machine-learning/inside-palantir-aip-how-the-worlds-most-controversial-ai-platform-actually-works

7. No Man's Hand: Artificial Intelligence Does Not Improve Police Report Writing Speed, https://www.crimrxiv.com/pub/nxbmzp2j

8. The Strategy And Tactics Of Pricing, http://www.pmm.edu.my/zxc/2022/lib/ebok/The%20Strategy%20And%20Tactics%20Of%20Pricing%20A%20Guide%20To%20Growing%20More%20Profitably%20By%20M%FCller,%20Georg%20Nagle,%20Thomas%20T.pdf

9. Data Collection Survey on Collaboration with Start-up Companies in Latin America and the Caribbean Final Report \- JICA Report PDF, https://openjicareport.jica.go.jp/pdf/12369658.pdf

10. Algorithmic Video Surveillance, dangers and counter-attacks \- La Quadrature du Net, https://www.laquadrature.net/en/avs/

11. How strategic is Chicago's Strategic Subjects List? | Upturn, https://www.upturn.org/work/how-strategic-is-chicagos-strategic-subjects-list/

12. CITY OF CHICAGO OFFICE OF INSPECTOR GENERAL ADVISORY CONCERNING THE CHICAGO POLICE DEPARTMENT'S PREDICTIVE RISK MODELS, https://igchicago.org/wp-content/uploads/2020/01/OIG-Advisory-Concerning-CPDs-Predictive-Risk-Models-.pdf

13. About Palantir, https://blog.palantir.com/about-palantir-ddddb78aec29

14. (PDF) Algorithmic Injustice: Predictive Policing and Profiling in the Netherlands \-the Top600 and Top400 approaches \- ResearchGate, https://www.researchgate.net/publication/388728682\_Algorithmic\_Injustice\_Predictive\_Policing\_and\_Profiling\_in\_the\_Netherlands\_-the\_Top600\_and\_Top400\_approaches

15. Top400 a top-down crime prevention strategy in Amsterdam \- Racism and Technology Center, https://racismandtechnology.center/wp-content/uploads/20221101-data-justice-lab-top400-report.pdf

16. AUTOMATING INJUSTICE: \- Police and Human Rights Resources, https://policehumanrightsresources.org/content/uploads/2021/09/Automating\_Injustice.pdf?x19059

17. EU AI Act Risk Classification Playbook: Prohibited vs High-Risk vs Limited-Risk, https://www.glocertinternational.com/resources/guides/eu-ai-act-risk-classification-playbook/

18. Partial ban on 'predictive' policing and crime prediction systems included in final EU AI Act, https://www.fairtrials.org/articles/news/partial-ban-on-predictive-policing-included-in-final-eu-ai-act/

19. Automated Apartheid: How Palestinians Became the World's Most Surveilled People, https://untoldmag.org/automated-apartheid-palestinians-surveillance/

20. Israel increasingly using facial recognition to track Palestinians \- Amnesty, https://www.timesofisrael.com/israel-increasingly-using-facial-recognition-to-track-palestinians-amnesty/

21. Predictive Policing Software Terrible At Predicting Crimes \- The Markup, https://themarkup.org/prediction-bias/2023/10/02/predictive-policing-software-terrible-at-predicting-crimes

22. Leading Indicators and Spatial Interactions: A Crime‐Forecasting Model for Proactive Police Deployment | Request PDF \- ResearchGate, https://www.researchgate.net/publication/227517779\_Leading\_Indicators\_and\_Spatial\_Interactions\_A\_Crime-Forecasting\_Model\_for\_Proactive\_Police\_Deployment

23. Good Policing Practices Are Difficult, Even for the Avengers \- EngagedScholarship@CSU, https://engagedscholarship.csuohio.edu/cgi/viewcontent.cgi?article=4265\&context=clevstlrev

24. An AI tool for helping police make custodial decisions, https://stopcorruption.ai/case-studies/ai-tool-for-helping-police-make-custodial-decisions/

25. Predictive Policing in Practice: A Case Study of Chicago's Strategic Subject List \- Scholarship @ Claremont, https://scholarship.claremont.edu/cgi/viewcontent.cgi?article=5092\&context=cmc\_theses

26. Zero-Inflated Embeddings to Analyze Homicide Occurrence Patterns \- WordPress.com, https://paularodrid.files.wordpress.com/2020/11/zero-inflated-embeddings-homicides.pdf

27. Person-Based Predictive Policing, https://www.nationalacademies.org/cdn/materials/9fba0812-55ec-4fae-971e-799acb07790c

28. AI in Courts and Sentencing \- PoliceAI News, https://policeainews.com/courts-sentencing.html

29. Machine Learning Algorithms and Police Decision-Making \- RUSI, https://static.rusi.org/201809\_whr\_3-18\_machine\_learning\_algorithms.pdf.pdf

30. European Travel Information and Authorisation System (ETIAS): Border management, fundamental rights and data protection, https://www.europarl.europa.eu/RegData/etudes/STUD/2017/583148/IPOL\_STU(2017)583148\_EN.pdf

31. Key takeaways for the private sector from The Bridges v South Wales police facial recognition case | Data Protection Report, https://www.dataprotectionreport.com/2020/08/key-takeaways-for-the-private-sector-from-the-bridges-v-south-wales-police-facial-recognition-case/

32. AI-X | CAS \- Crime anticipation system, https://ai-watch.github.io/AI-watch-T6-X/service/90080.html

33. trân trần xamvn, https://zenk-leuchtenvertretungen.de/amphtml/fr/analysis/video/rasi-sanyafimu

34. A Recommendation Engine to Aid in Identifying Crime Patterns | INFORMS Journal on Applied Analytics \- PubsOnLine, https://pubsonline.informs.org/doi/10.1287/inte.2019.0985

35. Artificial intelligence-related applications for railway security, https://uic.org/IMG/pdf/artificial\_intelligence-related\_applications\_web.pdf

36. 5 AI AND ADMINISTRATION OF CRIMINAL JUSTICE. REPORT ON THE NETHERLANDS By Maša GALIČ, Abhijit DAS and Marc SCHUILENBURG \* I. P, https://marcschuilenburg.nl/wp-content/uploads/2023/07/AI-AND-ADMINISTRATION-OF-CRIMINAL-JUSTICE.-REPORT-ON-THE-NETHERLANDS.pdf

37. Constraining Big Brother: The Legal Deficiencies Surrounding Chicago's Use of the Strategic Subject List, https://legal-forum.uchicago.edu/print-archive/constraining-big-brother-legal-deficiencies-surrounding-chicagos-use-strategic

38. Laplace's Officer: The Future of Predictive Policing \- CrimRxiv, https://www.crimrxiv.com/pub/laplace

39. Predictive Police Tech Isn't Making Communities Safer — It's Disempowering Them, https://truthout.org/audio/predictive-police-tech-isnt-making-communities-safer-its-disempowering-them/

40. Predictive Policing and the Constitutional Category Error \- Claremont Journal of Law and Public Policy, https://claremontlawjournal.sites.pomona.edu/?p=505

41. A Closer Look at Experian Big Data and Artificial Intelligence in Durham Police, https://bigbrotherwatch.org.uk/blog/a-closer-look-at-experian-big-data-and-artificial-intelligence-in-durham-police/

42. Algorithmic risk assessment policing models: lessons from the Durham HART model and 'Experimental' proportionality \- ResearchGate, https://www.researchgate.net/publication/324202083\_Algorithmic\_risk\_assessment\_policing\_models\_lessons\_from\_the\_Durham\_HART\_model\_and\_'Experimental'\_proportionality

43. Full article: Algorithmic risk assessment policing models: lessons from the Durham HART model and 'Experimental' proportionality \- Taylor & Francis, https://www.tandfonline.com/doi/full/10.1080/13600834.2018.1458455

44. Bridges v South Wales Police and emerging privacy concerns, https://nilq.qub.ac.uk/index.php/nilq/article/view/1127

45. Legal Challenge: Ed Bridges v South Wales Police \- Liberty, https://www.libertyhumanrights.org.uk/issue/legal-challenge-ed-bridges-v-south-wales-police/

46. UK Court of Appeal Finds Automated Facial Recognition Technology Unlawful in Bridges v South Wales Police \- Hunton Andrews Kurth LLP, https://www.hunton.com/privacy-and-cybersecurity-law-blog/uk-court-of-appeal-finds-automated-facial-recognition-technology-unlawful-in-bridges-v-south-wales-police

47. Fighting for algorithmic justice: lessons learned in working closely with affected people, https://edri.org/our-work/fighting-for-algorithmic-justice-lessons-learned-in-working-closely-with-affected-people/

48. Algorithmic Injustice: Predictive Policing and Profiling in the Netherlands – the Top600 and Top400 approaches \- Gupea, https://gupea.ub.gu.se/bitstreams/76e43905-85a8-4943-901a-9a3d02d52db3/download

49. Human Rights Implications of the Use of AI in the Digital Welfare State: Lessons Learned from the Dutch SyRI Case \- Oxford Academic, https://academic.oup.com/hrlr/article/22/2/ngac010/6568079

50. Predictively policed: the Dutch CAS case and its forerunners, https://repository.ubn.ru.nl/bitstream/handle/2066/289967/289967.pdf?sequence=1\&isAllowed=y

51. A Peer-Reviewed Newspaper, vol. 14, no. 1 (2025) FREE transmediale 2025, https://darc.au.dk/fileadmin/DARC/newspapers/Peer-reviewed-newspaper\_EDM-vol14\_2025.pdf

52. December 2020 ARTIFICIAL INTELLIGENCE AND CRIMINAL JUSTICE \- Caterina Chinnici – Europarlamentare, https://caterinachinnici.it/wp-content/uploads/2021/07/Copy\_Intelligenza-aritificiale-e-giustizia-penale-Copy.pdf

53. Smart Border is Watching You\! Fundamental Rights Implications of Automated Data Processing and Decision-Making at the EU Border \- Oxford Academic, https://academic.oup.com/book/58128/chapter/479900535

54. Crimes by source of information: SIEDCO is the official source of... \- ResearchGate, https://www.researchgate.net/figure/Crimes-by-source-of-information-SIEDCO-is-the-official-source-of-information-of-crimes\_fig3\_372314081

55. Modelling underreported spatio-temporal crime events \- PMC \- NIH, https://pmc.ncbi.nlm.nih.gov/articles/PMC10337961/

56. Algorithmic State Violence: Automated Surveillance and Palestinian Dispossession in Hebron's Old City \- Cambridge University Press & Assessment, https://www.cambridge.org/core/services/aop-cambridge-core/content/view/80B9C5192057ACEA17089E488CFC1486/S0020743823000879a.pdf/algorithmic-state-violence-automated-surveillance-and-palestinian-dispossession-in-hebrons-old-city.pdf

57. Israel/OPT: Israeli authorities are using facial recognition technology to entrench apartheid \- Amnesty International, https://www.amnesty.org/en/latest/news/2023/05/israel-opt-israeli-authorities-are-using-facial-recognition-technology-to-entrench-apartheid/

58. Al-Haq Submission on Artificial Intelligence Warns that Israel's AI-Driven Warfare and Surveillance are Erasing the Palestinian People's Right to Development, https://www.alhaq.org/advocacy/27054.html

59. Automated apartheid: Israel deploys facial recognition AI to 'supercharge segregation', https://www.peoplesworld.org/article/automated-apartheid-israel-deploys-facial-recognition-ai-to-supercharge-segregation/

60. Incident 672: 'Lavender' and 'The Gospel' AI Systems Reportedly Used in Gaza Targeting Operations with Civilian Harm Allegations, https://incidentdatabase.ai/cite/672/

61. The Use of the 'Lavender' in Gaza and the Law of Targeting: ai-Decision Support Systems and Facial Recognition Technology in \- Brill, https://brill.com/view/journals/ihls/16/2/article-p336\_003.xml?language=en

62. Vumacam wins 'spy camera' court case against city \- Moneyweb, https://www.moneyweb.co.za/news-fast-news/vumacam-wins-spy-camera-court-case-against-city/

63. THE HONG BAO CONNECTION \- Capital, https://www.capital-media.mu/2022/09/the-hong-bao-connection/

64. หลักการและทฤษฎี ความมั่นคงปลอดภัย \- DLP SSRU, https://ssrudlp.ssru.ac.th/data-file/topics/materials/20260202-213146-bff501b796af6ca50dad42bcda6f59d3-Unit%201%20Introduction%20to%20Security.pdf

65. AI Act | TechPolicy.Press, https://www.techpolicy.press/tracker/ai-act/

66. SoundThinking \- Wikipedia, https://en.wikipedia.org/wiki/SoundThinking

67. EU AI Act Summary: Everything You Need to Know in One Place \- eyreACT, https://eyreact.com/eu-ai-act-summary-everything-you-need-to-know-in-one-place/

68. Full article: Facial recognition on trial: data protection, discrimination, and the ethics of algorithmic governance in policing \- Taylor & Francis, https://www.tandfonline.com/doi/full/10.1080/13600834.2026.2638629

69. CPD's 'Heat List' and the Dilemma of Predictive Policing | RAND, https://www.rand.org/pubs/commentary/2016/09/cpds-heat-list-and-the-dilemma-of-predictive-policing.html