Belgian AI Burglary Prediction Project Sparks Privacy Debate
A Flemish research project using artificial intelligence to predict where and when burglaries are likely to occur has ignited a heated debate over the ethics of predictive policing, privacy protections, and whether such systems actually reduce crime. The project, known as BIGDATPOL (Big Data Policing), is led by Ghent University criminologist Prof. Wim Hardyns and funded with 2 million euros from the European Research Council, as RTBF reported.
How the System Works
The AI model divides participating police zones into 200x200 meter grid cells and analyzes approximately 70 environmental and socio-economic indicators to forecast burglary risk. These include the location of gyms, gas stations, hospitals, nightclubs, and hair salons, as well as street configuration, population density, income levels, unemployment rates, and even weather conditions. According to VRT NWS, the system also incorporates historical burglary data and demographic information.
“In areas with little street lighting, the chance of residential burglary is greater,” Hardyns told VRT NWS. “The presence of escape routes plays a major role. And weather is also important - even criminals prefer to stay home when it rains.”
The project explicitly does not aim to predict who will commit crimes - only where and when they are likely to occur. As Hardyns emphasized: “We focus on times and locations, with respect for privacy and human rights. That is central to our research.”
The model is currently undergoing a one-year field test across 19 Flemish police zones, following a promising pilot in the Zennevallei police zone. The BIGDATPOL project page describes the programme as addressing “one of the central challenges in European security: how to use big data and artificial intelligence responsibly and effectively for crime prevention.”
The Critics’ Concerns
The debate was sparked by a lengthy investigation published in the Flemish newspaper De Morgen on August 14, which raised serious questions about the project’s methodology and potential for bias. Researchers and anti-racist organizations have expressed particular concern about the inclusion of socio-economic data, including income levels, unemployment rates, and the proportion of residents of non-European origin.
Eli Verwimp, a researcher in algorithms at the Vrije Universiteit Brussel (VUB), warned in De Morgen: “If an algorithm detects a correlation between a variable and the result, it assigns it significant weight. But this correlation doesn’t necessarily exist in reality.”
Critics argue that a neighborhood that is more heavily policed may appear more often in crime statistics - not necessarily because it has more crime, but because more controls are conducted there. This mechanism could create a feedback loop, reinforcing police presence in already vulnerable or stigmatized neighborhoods.
A Cautionary Tale from the Netherlands
The debate comes at a time when the effectiveness of predictive policing systems remains unproven. The Netherlands quietly discontinued its comparable Criminaliteit Anticipatie Systeem at the end of 2025 after ten years of use, without clear demonstration of its added value. As NRC reported, the Dutch National Police abandoned the system that was designed to predict where and when crimes such as burglaries, street robberies, and pickpocketing might occur.
Similar systems have been discontinued elsewhere, including in Los Angeles, where the PredPol system was abandoned due to ethical concerns about disproportionate targeting of low-income neighborhoods.
The Project’s Defense
Prof. Hardyns and his team defend the project’s scientific rigor and ethical safeguards. In an in-depth interview with VVSG Magazine Lokaal, Hardyns emphasized that the project uses supervised machine learning - the system does not decide on its own, and there is always human oversight and interpretation.
The project team includes a legal expert and an ethics specialist, and training is planned for all involved police personnel. Hardyns also noted that the project complies with strict data protection, privacy, and AI-use regulations.
“Our ultimate goal is to use the police’s limited resources even better,” Hardyns said in the VVSG interview. He also expressed frustration with what he called “unfounded attacks by fellow scientists” about BIGDATPOL, stating that those statements “show a lack of interest in the actual content and results of our project.”
Why This Matters
The stakes extend well beyond Belgium. If successful, the model could be rolled out across Europe. According to the UGent TechTransfer, exploratory talks are already underway with Europol and countries including Serbia, Spain, the Netherlands, Hungary, Switzerland, and Austria. The project was also presented at the Burglary Prevention Seminar 2026 in Leuven, organized by AGORIA and ALIA Security.
The debate also comes at a moment when burglary numbers in Belgium are actually declining - since 2020, they barely exceed 40,000 per year, compared to approximately 56,000 in 2016.
Academic research on the project’s methodology continues to be published. A June 2025 paper by Robin Khalfa and Wim Hardyns, published in Applied Spatial Analysis and Policy, compared machine learning-based crime predictions across different micro-geographic units, finding that grid-based approaches offer a balanced performance for predicting crime risks.
What’s Next
BIGDATPOL must now convince on two fronts: its concrete utility for police officers and its ability to avoid discrimination or territorial stigmatization. The one-year field test across 19 police zones will be crucial in determining whether the system delivers measurable results.
Key questions remain unanswered: Will the field test demonstrate a meaningful reduction in burglaries? How will the project address concerns about the inclusion of ethnic origin data? What safeguards will prevent the feedback loop of increased policing leading to more recorded crime in the same areas? And how will the project respond to the Dutch experience, where a similar system was abandoned after a decade?
As the debate unfolds, the fundamental tension at the heart of modern policing remains unresolved: the promise of using big data and AI to make policing more efficient and proactive, versus the risks of algorithmic bias, discrimination, and the erosion of public trust in already marginalized communities.