Law360 Canada (July 23, 2026, 12:58 PM EDT) --

Aavisa Butt

The accelerating development of artificial intelligence, both domestically and internationally, provides important lessons for the responsible implementation of predictive policing systems within criminal justice frameworks. This is particularly evident in the governance challenges associated with algorithmic policing technologies, including regulatory lag, surveillance arbitration, symbolic compliance and weak oversight.

While Canadian police services continue to deploy algorithmic policing technologies and predictive tools, the absence of formal, transparent, government-led audits assessing their effects on bias and racial discrimination represents a significant governance gap.

As stated by Shawn Singh in his 2022 Manitoba Law Journal article on algorithmic policing technologies (APTs), “serious deleterious potential for the Charter rights of Canadians, with consequences that disproportionately affect people of colour. Be that as it may, the most malevolent consequence of applying APTs may be their application of generalized formulas to generate recommendations used to intercept individuals based on biased and inaccurate information. Although not authorized by statute, surveillance technologies continue to be permissible under common law authorities.”

By examining the Los Angeles Police Department’s use of PredPol (now Geolitica) — scrutinized via official Office of the Inspector General (OIG) audits — we can contrast vendor marketing claims with how predictive algorithms actually operate. Because Canada shares a heavy reliance on North American tech providers, the U.S. trajectory serves as a crucial cautionary tale. By learning from American oversight failures, Canada can proactively shield its citizens from systemic algorithmic flaws. Ultimately, we must prioritize the rule of law over corporate incentives through robust institutional compliance audits and public accountability.

We will evaluate PredPol as a socio-technical system, revealing its asymmetrical social power dynamics, which are further exacerbated once this technology enters the criminal justice system and court processes. Its effectiveness in reducing crime has not been thoroughly proven and remains contested. The historical crime data it utilized revealed concerns of “dirty data,” establishing itself as a vulnerable yet biased tool, giving rise to further structural inequalities and racial discrimination embedded within societies.

In the 2010 article Predictive Policing: The Future of Law Enforcement?, Beth Pearsall defines predictive policing as “taking data from disparate sources, analyzing them and then using results to anticipate, prevent and respond more effectively to future crime.” It predicts the time and location where crimes are likely to occur.

Drawing on direct interviews with PredPol, The Markup’s 2021 investigation reported that “to use PredPol’s algorithm, police departments set up an automatic feed of crime reports, which experts and police said include incidents reported by both the public and by officers, and choose which crimes they want predicted.”

Here, the input appears to be historical crime data: type of crime, location, time. There is pattern identification using machine learning models. Whereas the output appears to be geospatial hotspots: maps visually depicting where crime is most likely to occur next, thus recommending that police officers patrol there.

The official audit, conducted by the Office of the Inspector General of the Los Angeles Police Commission, stated that PredPol predicts where crime will most likely occur within “the next 12 hours,” adding that it “analyzes 10 years of crime data” and that “results are generated by the software’s algorithm.”

Moreover, following the discontinuation of this AI, the official website of the City of Los Angeles reported that the “crime data indicated a decrease of 19 per cent in homicides in 2025, compared to 2024. The number of homicides for 2025 is the lowest since 1966.”

PredPol “dosage” refers to the amount of time police spend in areas that PredPol identifies as high risk for crime. The OIG audit revealed that the data, using two dashboards, did not match, often by a large margin, one by Palantir, the other by PredPol’s automatic measurement. Manual logging depended on officers inputting their hours themselves, whereas automatic logging (though with inconsistencies) read the GPS and whether cars were near police stations. The report discovered that though police spent “41 per cent more time” in these predicted areas from 2017 to 2018, vehicle-related crime barely changed, with only a three per cent drop. Most places were visited very few times, and the visits themselves were quite short.

There is also an economic rationale underlying predictive policing. In their 2024 article, Algorithmic Crime Prevention: From Abstract Police to Precision Policing, Simon Egbert and Elena Esposito argue that “police should focus their resources on those people and places, which are most closely connected to crime risks.” They further emphasize efficiency, explaining that “thus, contextual reference (location and time) … lies not in the ‘predictive side’ but in ‘patrol operations management,’ which should increase efficiency in keeping the community safe.”

This form of resource allocation is what advocates for this technology repeatedly emphasize, instilling the narrative in people that due to the lack of police personnel, who work in favour of public safety, AI systems should be repeatedly deployed in the interests of citizens. However, this presumption hinges on the idea that simply due to increased police presence, crime deterrence is effective. This is an ongoing debate within the academic sphere.

However, effectiveness alone is not sufficient justification. Even if crime declines, predictive policing must still be evaluated for bias, constitutional rights, legitimacy and governance. A Courthouse News Service report stated that “data on 484,000 pedestrians stopped and questioned by LAPD officers between July 2012 and June 2014 revealed that officers stopped black and Latino residents 33 per cent and 46 per cent of the time, respectively, while white residents were only stopped 17 per cent of the time, according to the report.”

But this is the same data, embedded with historical and structural inequalities, being fed into predictive policing systems. This is where the “neutral data” problem occurs. Predictive policing is machine learning. It is a subset of artificial intelligence where computers learn from data and identify patterns without being explicitly programmed. Instead of following hard-coded instructions, algorithms analyze large datasets to make accurate predictions, classify information or generate new content, continuously improving their performance over time.

In other words, what you feed it is what it will learn from.

In its defence, PredPol claimed it did not use race or socioeconomic data. However, it did use crime location data. And crime location data reflects sites of historical over-policing and enforcement bias, leading to unequal state surveillance. Simply because “racial” data is not encoded into the algorithm does not mean it cannot produce directives, such as to visit a certain location with a higher indication of crime potentially occurring, that do not already have racial bias within them. As demonstrated in the 2019 New York University Law Review article, “Dirty Data, Bad Predictions: How Civil Rights Violations Impact Police Data, Predictive Policing Systems, and Justice,” historical police data is often “dirty,” meaning that it may reflect and reproduce prior civil rights violations. This can contribute to feedback loops, whereby predictive policing systems reinforce and perpetuate the same patterns embedded within the underlying data.

This practice potentially threatens equality rights guaranteed under s. 15(1) of the Canadian Charter of Rights and Freedoms, as systemic feedback loops are masked under the guise of technological neutrality. Consequently, traditional legal definitions of discrimination are ill-equipped to capture or evaluate the hidden biases inherent in algorithmic governance.

Furthermore, this approach drives immediate arrests and intensive police patrolling, resulting in severe state interference with privacy — tactics that prosecutors routinely defend in court as a necessary “rationale” for public safety. With this, when we consider how artificial intelligence does not always maintain its reasoning as to why it delivered a certain decision, we encounter the “black box” problem. In the chapter, “Human Rights and Algorithmic Impact Assessment for Predictive Policing” in Constitutional Challenges in the Algorithmic Society, Céline Castets-Renard explains, “An explanation as to why a model has generated a particular output or decision (and what combination of input factors contributed to that) is not always possible. These cases are referred to as ‘black box’ algorithms and require special attention.”

Therefore, while the Canadian Charter of Rights and Freedoms demands respect for privacy, suspicion itself becomes rewritten. If arrests can be attributed to a mysterious, faulty decision from the technological void of AI, privacy is threatened by discriminatory data, and the right to a fair trial is undermined because these “black box” decisions cannot be readily challenged in court. Proprietary code and training data are often shielded as trade secrets.

Legally, one is innocent until proven guilty. However, PredPol subverts this by reconfiguring suspicion itself. Suddenly, anyone and everyone could be a suspect because this technology surveils everyone in advance of individualized suspicion. This raises a fundamental constitutional question regarding state interference with privacy: is predicting future crime a proportionate justification for mass pre-emptive surveillance?

As Castets-Renard explains, “It reduces the complexity of reality by calculating the probability that certain facts happen on the basis of repeated actions.”

PredPol challenges the presumption of innocence by transforming statistical probability into reasonable suspicion, establishing a dangerous legal precedent where algorithmic prediction replaces the concrete factual evidence traditionally required after an arrest. While critics might argue that this simply reflects existing crime data, that defence collapses: the system merely creates a self-fulfilling loop that uses biased policing to validate its own dirty data.

Stakeholder responses further demonstrate the contested nature of predictive policing technologies. A 2019 MuckRock investigation reported that one Los Angeles resident criticized PredPol’s increased police presence, describing it as “a program that has unjustly targeted and stalked people.” Similarly, Santa Cruz became the first U.S. city to ban predictive policing technologies amid broader concerns regarding algorithmic bias and racial disparities. As reported by the Thomson Reuters Foundation in 2020, Santa Cruz Mayor Justin Cummings stated, “Understanding how predictive policing and facial recognition can be disproportionately biased against people of colour, we officially banned the use of these technologies in the city of Santa Cruz.”

Above all, effective policing requires recognizing that algorithms are not a universal solution. The LAPD case underscores the vital need for rigorous audits, particularly since the software was abandoned following its investigation. Because these models risk reproducing structural inequalities, data inputs and design must be critically interrogated. For Canada, these lessons are urgent: as police services adopt similar tools, the absence of formal, government-led audits leaves a governance gap. By moving past vendor claims of neutrality, Canada must institute robust oversight to safeguard Charter rights and prevent the systemic harms seen in the U.S.

Aavisa Butt is pursuing a master’s in artificial intelligence with a specialization in criminal justice systems. She holds an honours bachelor’s degree in English literature and an honours paralegal diploma.

The opinions expressed are those of the author(s) and do not necessarily reflect the views of the author’s firm, its clients, Law360 Canada, LexisNexis Canada or any of its or their respective affiliates. This article is for general information purposes and is not intended to be and should not be taken as legal advice.

Interested in writing for us? To learn more about how you can add your voice to Law360 Canada, contact Analysis Editor Peter Carter at peter.carter@lexisnexis.ca or call 647-776-6740.