Law enforcement and Artificial Intelligence: the state of a transformation already under way | Stud
From ShotSpotter’s 87% false alarm rate in Chicago to the 98% accuracy of the Polizia di Stato’s system for detecting child sexual abuse material: the Eurispes study on the use of AI by law enforcement compares the models of the United States, China, India, the United Kingdom and the European Union with the situation in Italy, amid systems already in operation, more than 143 million euros of investment under way and questions still open on transparency and accountability.
The Eurispes study devoted to the use of Artificial Intelligence by law enforcement sets out the theoretical framework of the phenomenon, analyses the main international experiences (United States, China, India, United Kingdom, European Union) and looks in specific depth at the situation in Italy, before addressing the ethical and governance issues associated with the use of these systems.
A capability multiplier, not a mere tool
Unlike previous technological revolutions in public security, from radio communications to electronic databases, AI does not simply support the work of law enforcement but redefines its operational logic: from prevention to prediction, from reaction to proactive surveillance. The applications analysed range from predictive analytics to biometric recognition, from natural language processing to digital forensics, through to the analysis of criminal and financial networks.
Mixed results in the international comparison
The study documents operational results that are in some cases very positive – such as the 6% reduction in crime in New York with the ShotSpotter system integrated with drones, or the 7% reduction in Los Angeles with area-based predictive analytics – alongside spectacular failures: in Plainfield (New Jersey) the Geolitica system produced more than 23,000 predictions with a success rate of less than 0.5%, while a 2024 audit found that 87% of ShotSpotter alerts in Chicago did not correspond to actual shootings. The COMPAS system, used in American courts to assess the risk of reoffending, has been criticised for its disproportionately unfavourable classification of African American defendants.
China represents the most systematic case of integration, with infrastructures such as Skynet and Sharp Eyes and platforms such as Zhejiang’s Police Cloud, which in 2019 was handling more than 1.6 trillion data points. The United Kingdom, although it has no dedicated legislation, has developed a more structured system of safeguards; the European Union, with the AI Act, classifies almost all the systems used by law enforcement as “high-risk”, imposing stringent obligations of transparency and human oversight.
The Italian case: robust investment and an active safeguard
In Italy, the Polizia di Stato’s SARI system has already produced positive investigative results in its Enterprise component, while the Real-Time version was blocked by the Italian Data Protection Authority in 2021 for lack of an adequate legal basis. The X-Law system, trialled in Naples and adopted by several provincial police headquarters, has shown reliability rates of between 87% and 97%. The Polizia di Stato’s Centre for Research and Analysis of Multimedia Information (CRAIM) achieved 98% accuracy in the testing phase of a system for recognising child sexual abuse material, with the further aim of reducing investigators’ exposure to sensitive content.
The Internal Security Fund 2021-2027 supports a portfolio of more than 40 projects in Italy with a total value in excess of 143 million euros, including CEREBRO 2 (Polizia di Stato, more than 27 million euros) to combat illicit assets.
Responsible governance of AI
The study identifies a number of priorities for the responsible governance of AI in public security, including: clear and traceable chains of accountability for every decision taken with algorithmic support; mandatory and regular impact assessments; closing the training gap between operators and the technical complexity of the systems; independent verification of the performance claimed by suppliers before large-scale adoption; and genuine involvement of citizens, through open registers and mechanisms for challenging decisions, as a condition of democratic legitimacy.
Artificial Intelligence should be regarded neither as a threat to be contained nor as a technical solution to be adopted without reservation, but as a radical transformation that requires, for Italy too, institutional investment in clear regulatory frameworks, sound assessment tools and organisational cultures equal to its complexity.
