The mafia has long been recognized as one of the most resilient forms of organized crime, capable of surviving generational change and adapting to shifting social, economic, and technological contexts. Giovanni Falcone captured this enduring adaptability when he wrote: “Mafia is characterized by its ability to rapidly adapt archaic values to the needs of the present, by its skill in blending into civil society, by its use of intimidation and violence, by the number and criminal stature of its members, and by its capacity to be always different and yet always the same. We must destroy the myth of the so-called new mafia or, rather, we must convince ourselves that there is always a new mafia ready to replace the old one.”[1]

Today, the adaptive capacity of mafia organizations finds its most striking expression in their strategic adoption of digital tools and artificial intelligence (AI). This shift is not merely the result of technological evolution but is primarily driven by the need to evade law enforcement controls and reduce the risk of interception. Thanks to vast and liquid financial resources, criminal networks can attract top experts in computer science, blockchain, and AI, securing access to cutting-edge technologies often before state authorities themselves. In this way, they can manage digital wallets, launder illicit revenues, and communicate through encrypted platforms that guarantee higher levels of anonymity.

AI further amplifies these capabilities by automating operations such as transaction anonymization, data obfuscation, and even cyberattacks. Combined with cryptocurrencies, encryption systems, and the dark web, AI allows criminal organizations to lower operational costs, increase transaction security, and expand their activities on a global scale. A well-documented example is the 2015 Antwerp case, in which IT engineers linked to the ’Ndrangheta manipulated port systems to facilitate large-scale drug shipments.

These technological innovations therefore serve a dual purpose: enabling mafias to keep pace with global digital transformations while simultaneously widening the gap with law enforcement. As a result, investigations become increasingly complex and fragmented, while AI strengthens the ability of organized crime groups to operate undetected. This highlights the urgent need for a more coherent and coordinated system of global governance capable of rebalancing this structural asymmetry.

At the same time, geopolitical crises and institutional instability in regions from Latin America to West Africa have fostered the expansion of criminal networks. Today’s drug trafficking routes pass through hubs such as Brazil, Venezuela, and Ecuador, with outlets in Northern European and Mediterranean ports, while new pathways are consolidating in the Sahel and West Africa. These systems operate as genuine global holdings, relying on transnational brokers, complex logistics, and parallel financial channels, often with the complicity of local operators.

The case of Raffaele Imperiale, a former affiliate of the Italian mafia and widely regarded as one of the most prominent figures in global drug trafficking, exemplifies this evolution. Arrested in Dubai in 2021 and later becoming a state witness, Imperiale coordinated a transnational network operating between South America and Europe. His organization relied on alternative money transfer systems such as hawala, speculative investments in gold, and encrypted communication platforms like EncroChat, highlighting the increasing convergence of traditional criminal structures with sophisticated financial mechanisms and digital technologies. Operation Emma 95 revealed how such networks are structured in a reticular and decentralized manner, composed of autonomous yet interconnected units. Contemporary mafia organizations no longer simply replicate past models but function as highly adaptive global enterprises.[2]

On the other hand, since the 1960s and 1970s, investigators had already understood that mafia-related crimes could not be read as isolated episodes but rather as part of a broader strategic design. This insight led to the development of the so-called “anti-crime method,” based on genealogical reconstruction, contact mapping, and systematic observation of behaviors.[3] Yet, the lack of technological tools made these practices extremely costly in terms of time and resources. Today, advances in digital and computational technologies have revolutionized such approaches, making it possible to integrate traditional investigative knowledge with tools capable of analyzing vast amounts of data and providing complex information syntheses to guide strategic countermeasures.

This study examines how AI is reshaping investigations into organized crime, highlighting not only the operational opportunities and emerging risks (1) but also the concrete applications and projects that demonstrate its impact in practice (2).

The Investigative Phase

Among the areas where the use of AI-based applications has grown most significantly is the investigative phase, where intelligent systems serve as valuable support tools for law enforcement activities (A). These technologies accelerate investigations and enhance operational efficiency, while also raising important challenges concerning privacy and the protection of personal data (B).

A. Predictive AI 

Predictive AI represents the category of AI that “looks to the future” and is applied in two main areas: predictive justice and predictive policing.

In predictive justice, we find both defensive predictive algorithms, which analyze large volumes of judicial rulings to estimate the likely outcome of a case and guide the litigation strategies of the parties, and decision-oriented predictive algorithms, used by judges or other authorities to assess risk profiles. These include risk assessment tools employed to estimate the probability of recidivism and to inform decisions on pre-trial detention, security measures, or conditional suspension of sentences.

Predictive policing, on the other hand, is based on statistical forecasts.[4] By analyzing historical crime data, models can be built to estimate the probability of a given crime occurring in a specific place or time [place-based systems (i)], or to identify individuals at higher risk of criminalization or victimization [person-based systems (ii)]. Predictive policing thus marks a crucial shift from retrospective analysis to proactive prevention of criminal activity.

Operationally, the process unfolds in three main phases:[5]

• Data collection: Historical information on crimes (type, location, time, methods) is gathered, and in person-based systems, personal data such as ethnicity, education level, or socioeconomic conditions are also included.
• Data analysis: Algorithms process these inputs and generate risk scores (low, medium, high).
• Police operations: The results are used to guide operational decisions, such as increasing law enforcement presence in specific areas or monitoring individuals with a higher likelihood of reoffending.[6]

The added value of such systems is evident, given the complexity of modern investigations. AI helps overcome the risk of “information overload” by instantly detecting connections among heterogeneous elements. These tools clearly enhance the effectiveness of investigations, making analyses more accurate, optimizing resource allocation, and reducing both time and costs.

For example, forecasting where and when a crime is most likely to occur allows law enforcement to be present in those areas at the predicted times, discouraging criminal activity or enabling rapid intervention.

At the same time, however, as will be explored more thoroughly in the following section (2), these systems raise serious concerns regarding respect for fundamental rights and individual freedoms.

I. Place-Based Systems

• Identifying Risk Areas

Crime is not evenly distributed across a territory; it tends to cluster in “hot spots.” By analyzing these patterns, it becomes possible to predict where similar offenses are likely to reoccur.[7] Some methods operate much like tracking an epidemic: if a burglary takes place in one area, similar incidents are likely to follow nearby within a short time frame. This so-called “near-repeat effect”[8] inspired the software PredPol, which, with just three data points (type of crime, location, and time), generates maps of the areas most at risk. Field tests showed significant reductions in theft and robbery rates, demonstrating the effectiveness of predictive approaches.[9]

• Forecasting Time and Place Together

Other systems go a step further by predicting not only where but also when crimes are likely to occur. In Italy, the software X-LAW processes in real time police reports, patrol data, and socioeconomic indicators, updating a digital map every thirty minutes with the areas and time slots at greatest risk. This allows law enforcement to deploy resources more precisely.[10]

• Explaining Why Certain Areas Are Vulnerable

Some tools aim to identify the structural causes that make an area more exposed to crime: poor lighting, abandoned buildings, or proximity to transport hubs can all increase vulnerability. The RTMDx software integrates environmental and urban planning data with crime statistics, producing risk maps that help design long-term prevention strategies.[11]

II. Person-Based Systems

• Risk Profiling: Learning from the Past

Risk profiling is one of the most established techniques in predictive justice. It is based on the idea that by observing recurring traits among large populations of offenders, it becomes possible to identify the most relevant risk factors, such as age, prior convictions, family history, or socioeconomic conditions. When an individual displays characteristics similar to those of persistent or violent offenders, they are considered more likely to reoffend. This methodology is often applied during judicial proceedings, where risk assessment tools support judges in decisions on preventive detention, conditional release, or other precautionary measures, thereby reducing the subjectivity of such evaluations.[12]

• Mapping Criminal Networks

Organized crime does not operate chaotically; it is built on networks of trust, exchanges, and intermediaries. Social Network Analysis (SNA) maps these relationships, showing not only who has the most connections but also who links otherwise separate groups. These intermediaries, more than visible bosses, often ensure the cohesion of the organization. Targeting them with focused arrests can destabilize the network far more than removing formal leaders, who are typically easier to replace. Empirical evidence supports this: studies of Sicilian mafia networks have shown that eliminating just 5% of the most strategic intermediaries reduced overall connectivity by as much as 70%, effectively fragmenting the organization. However, resilience does not depend solely on social ties but also on individual expertise. A chemist specialized in producing synthetic drugs, for instance, may be indispensable even without a central communication role. For this reason, the most effective strategy is to combine the identification of key intermediaries with that of members possessing irreplaceable skills.[13]

B. Technical Limitations of AI and Comparative Analysis of Regulatory Frameworks

AI has the potential to transform investigations by supporting prevention, detection, and analysis of criminal activity. Yet its deployment introduces crucial risks.

I. Potential Risks

The first is the illusion of neutrality. Every AI system reflects the choices and limitations of its programmers, from the selection of data to the weight assigned to variables. These embedded biases can reinforce inequities or produce discriminatory outcomes.[14]

A second limitation is opacity. Many advanced AI models, especially deep learning architectures, operate as “black boxes,” making it difficult to trace how they reach their conclusions. This undermines public trust, complicates bias correction, and raises accountability questions: is the responsibility borne by the developer, the user, or the deploying institution?

In criminal justice, these risks are magnified. Predictive policing and investigative profiling often rely on personal data that can conflict with the right to privacy and the presumption of innocence. Systems that flag individuals as “high risk” based solely on statistical correlations risk clashing with democratic principles. The European Union’s AI Act explicitly prohibits judgments based purely on personal traits or statistical data without verifiable evidence—highlighting the tension between efficiency and fundamental rights.[15]

II. Comparative Perspectives on AI Regulation

The governance of AI differs sharply across global powers, shaping both its potential to serve justice and its vulnerabilities to criminal misuse. These differences not only affect the development and adoption of technology but also directly shape the opportunities for organized crime to exploit AI and the prospects for international cooperation.

• Europe: The Regulatory Model

The European Union has adopted a strongly regulatory approach, based on the conviction that innovation and the protection of fundamental rights must advance together to generate public trust. Its legal framework rests on consolidated pillars such as the GDPR and the Law Enforcement Directive, to which the AI Act—introduced in 2024 as the world’s first comprehensive regulation dedicated to AI—was added.

The AI Act establishes a risk-based model: certain applications are prohibited because they are considered excessively intrusive (for example, real-time biometric surveillance); others, classified as “high risk,”          are subject to stringent requirements of transparency, accuracy, human oversight, and impact assessments on fundamental rights; while low-risk applications can develop more freely within general legal boundaries. This structure seeks both to limit abuse—making it more difficult for criminal organizations to exploit AI in activities such as money laundering or disinformation—and to reinforce citizen confidence.

However, the European approach is not without limits. The high level of regulation risks incentivizing criminal groups to relocate their activities to jurisdictions with weaker oversight, exploiting the fragmentation of global governance.[16]

• United States: The Self-Regulatory Model

The U.S. model has historically been rooted in self-regulation, consistent with its market-driven tradition and the philosophy of “move fast, break things first, apologize later.” This approach has allowed the United States to maintain a position of global leadership in AI development, thanks to early public investment and the dynamism of the private sector.

The absence of binding regulation, however, has a dual effect. On the one hand, it fosters a fertile environment for technological innovation; on the other, it creates vulnerabilities that can be exploited by organized crime. Practices such as automated phishing, deepfakes, or cryptocurrency laundering flourish more easily in a system lacking clear national standards on ethics, transparency, and data protection. While U.S. law enforcement agencies possess advanced tools to counter these threats, the absence of uniform regulation leaves gaps and opportunities for abuse.[17]

• China: The Authoritarian Regulatory Model

China presents a profoundly different model, characterized by massive public investment and centralized political control. The Next Generation AI Development Plan, launched in 2017, set the objective of achieving global leadership by 2030. On the regulatory side, Beijing distinguished itself as the first country in the world to adopt specific rules on generative AI (July 2023). These regulations mandate algorithm registration, data traceability, preventive compliance checks, and adherence to “socialist values.”

This approach drastically reduces the room for maneuver of local mafias: pervasive surveillance limits the use of AI for fraud, trafficking, or money laundering. However, it pushes criminal organizations to move their operations abroad, exploiting jurisdictions with weaker regulation. The Chinese model also raises two significant concerns: domestically, the fragile boundary between crime prevention and political repression; internationally, limited transparency and poor compatibility with global standards, which complicate multilateral cooperation.[18]

• The Global Cost of Fragmented AI Governance

A fragmented governance system produces consequences that go far beyond domestic vulnerabilities. Regulatory incoherence between Europe, the United States, and China not only creates loopholes for criminal groups at the national level but also severely undermines international coordination.

First, cybercriminals exploit these discrepancies through forum shopping: they strategically locate servers, cryptocurrency wallets, or command-and-control infrastructures in jurisdictions with weaker regulations or fewer cooperation requirements. This makes it extremely difficult for law enforcement to trace payments, assign responsibility, or seize illicit assets, as mutual legal assistance requests often stall due to incompatible legal frameworks.

Second, the absence of shared standards for data management and digital evidence slows down investigations. For instance, evidence collected under GDPR rules in Europe may not be admissible in U.S. courts, while in China, it may be entirely inaccessible due to restrictions on cross-border data sharing.

Third, diverging judicial timelines exacerbate the problem: cybercrimes are executed and spread almost instantly, but institutional responses often take weeks or months. This delay gives criminal groups ample time to erase traces, move funds, or reorganize their networks.

Attempts to build convergence exist—the Budapest Convention on Cybercrime remains the only binding international treaty in this field—but the absence of key actors such as China and Russia undermines its global reach.

The result is a paradox: while organized crime seamlessly exploits the transnational nature of cyberspace, authorities remain confined within national jurisdictions that do not always communicate effectively. This asymmetry makes every operational success fragile and temporary, highlighting the urgent need for a more coherent and interoperable global framework.[19]

Strategic and Operational Uses of AI

Organized crime, by its transnational and adaptive nature, exploits technological vulnerabilities and regulatory gaps; consequently, the use of advanced AI solutions becomes essential to reduce the asymmetry between criminal and state capabilities. Yet this scenario is still in the process of consolidation: current applications demonstrate significant potential but remain marked by technical, operational, and legal limitations that require critical assessment. The following analysis, therefore, focuses on concrete experiences, highlighting how AI has already contributed to dismantling criminal networks, understanding mafia recruitment processes, and strengthening mechanisms of international cooperation.

A. ENCROCHAT: decoding the invisible

One of the most striking cases showcasing the potential of advanced data analytics and AI tools in investigations came with the dismantling of EncroChat in 2020 through an international collaboration coordinated by Europol.

The platform, used by around 60,000 users worldwide, had been designed to guarantee anonymous and untraceable communications, making it a preferred tool for criminal organizations engaged in drug trafficking, money laundering, extortion, and even murder. The infiltration and subsequent decryption of EncroChat enabled law enforcement agencies to intercept over 115 million encrypted messages, exposing in real time the planning of criminal operations.

The operational impact was extraordinary: data analysis led to 6,558 arrests globally, including 197 high-value targets, as well as the seizure of large quantities of narcotics, weapons, and luxury goods. The decryption of communications also made it possible to identify recurring patterns and strategic nodes within criminal networks, providing investigators with a detailed map of internal clan connections and intermediation dynamics.

The EncroChat case highlights at least three crucial aspects for the future of investigations against organized crime. First, it demonstrates the effectiveness of supranational investigative alliances, which are essential to tackle criminal phenomena that operate across national borders. Second, it underscores the strategic role of advanced data analytics technologies, which allow massive volumes of heterogeneous information to be transformed into actionable intelligence. Finally, it shows how the ability to penetrate encrypted digital platforms can radically shift the balance of power between criminal organizations and law enforcement, reducing the anonymity on which the former relies.[20]

B. PROTON: preventing recruitment before it happens

If EncroChat exposed existing networks, Proton represents a proactive application of AI: understanding how criminal groups recruit. Developed by the research center Transcrime and funded by the European Commission, Proton combines social science insights with virtual simulations. By recreating digital communities, it allows policymakers to test how different interventions might affect the likelihood of individuals joining criminal organizations. A pilot in Palermo compared a “top-down” strategy of tougher policing with a “bottom-up” campaign of civic awareness. The simulations demonstrated how these approaches produce diverging long-term effects, helping decision-makers strike a balance between deterrence and community empowerment. Proton highlights AI’s potential not only for investigation but also for prevention, offering a lens into the social dynamics that sustain organized crime.[21]

C. I – CAN: international cooperation in practice

The ’Ndrangheta today stands as one of the most powerful and pervasive criminal organizations in the world, capable of combining deep territorial roots with a global vocation. Through privileged ties with Latin American drug producers, it has evolved into a true transnational “holding,” with interests ranging from drug trafficking to money laundering and the control of public contracts. In this context, the I-CAN initiative (Interpol Cooperation Against ’Ndrangheta), launched by Italy together with Interpol, represents an innovative model of multilateral cooperation.

The initiative rests on three key pillars: the valorization of content (namely, Italy’s long-standing expertise in the fight against the mafia), access to vital data and information for participating countries, and joint action through coordinated international operations. This structure has made it possible to systematize fragmented knowledge, centralize intelligence, and strengthen operational cooperation among police forces across continents.

The results are tangible: more than 150 arrests, including high-profile figures such as Edgardo Greco, a ’Ndrangheta fugitive who was on the run for nearly 17 years and sentenced to life imprisonment for a double murder in 1991, and Rocco Morabito, a notorious international drug trafficker and key member of the ’Ndrangheta criminal organization, whose capture represents a major blow to organized crime. In addition, Operation “Eureka” dismantled criminal networks operating across several European countries. The innovative aspect of the initiative lies in the use of advanced analytical tools and the creation of a strategic hub within Italy’s Central Directorate of the Criminal Police, capable of interfacing with both national and international databases.[22]

Conclusion

AI has become an indispensable tool in the fight against organized crime, reshaping the investigative landscape by enabling law enforcement agencies to process data on a scale and speed previously unimaginable. From encrypted communications to financial records, AI supports both preventive and repressive strategies, enhancing the capacity to anticipate risks, identify hidden connections, and dismantle criminal networks.

Its greatest strength lies in complementing human expertise. Algorithms can filter vast amounts of information, highlight correlations, and reduce investigative complexity, but it is the investigators who contextualize results, validate leads, and decide on strategic actions. AI, therefore, should not be seen as a substitute for human intelligence but as a force multiplier that frees resources and strengthens decision-making.

Nevertheless, the integration of AI into investigations raises crucial ethical and legal challenges. Predictive policing and algorithmic profiling, if not subject to independent oversight and quality data controls, risk consolidating social inequalities rather than reducing crime. Far from being neutral, algorithms reflect the biases of their designers and the limitations of the data on which they are trained. Without transparency and accountability, these systems may erode public trust—the very foundation of their legitimacy.

Comparative perspectives highlight how governance choices shape both opportunities and vulnerabilities. The European Union prioritizes rights and trust through strict regulation but risks slowing innovation. The United States favors innovation and market leadership, yet often underestimates risks linked to privacy and accountability. China couples massive investment with centralized control, limiting domestic misuse but bending AI to political objectives. None of these models is sufficient alone. For AI to serve justice while mitigating risks, international cooperation is essential to harmonize standards and close regulatory gaps exploited by transnational criminal groups.

Ultimately, the question is not whether AI will be used in investigations, but how it will be used. If deployed responsibly, with clear safeguards and international alignment, AI can reinforce democratic values, build trust, and give law enforcement the tools needed to confront one of the most adaptive threats of our time: organized crime.

[1] Giovanni Falcone and Marcelle Padovani, Cose di Cosa Nostra (Milano: Rizzoli, 1991).

[2] Nicola Gratteri and Antonio Nicaso, One and the Same: How the Mafias Have Integrated with Power (Milan: Mondadori, 2025).

[3] Ivan Rizzolo, “The General’s Lesson,” Supplement no. 3 of the Rassegna dell’Arma dei Carabinieri, no. 3, 2022.

[4] Fabio Basile, Artificial Intelligence and Criminal Law: Some Updates and New Reflections.

[5] Camaldo Lucio Bruno Cristiano, Artificial Intelligence and Predictive Criminal Investigation.

[6] Fabio Basile, Artificial Intelligence and Criminal Law: Some Updates and New Reflections.

[7] Beatrice Perego, “Predictive Policing: Algorithm Transparency, Impact on Privacy, and Discriminatory Implications.”

[8] Andrew Guthrie Ferguson, The Rise of Big Data Policing: Surveillance, Race, and the Future of Law Enforcement (New York: New York University Press, 2017).

[9] Robert Mitchell, “Predictive Policing Gets Personal,” Computerworld, October 24, 2013, https://www.computerworld.com/article/1603273/predictive-policing-gets-personal.html.

[10] Claudia Morelli, “X-Law, il brevetto italiano di polizia predittiva,” Altalex, November 7, 2022, https://www.altalex.com.

[11] Risk Terrain Modeling, https://www.riskterrainmodeling.com/.

[12] National Institute of Justice, “Best Practices for Improving the Use of Criminal Justice Risk Assessments,” 2024, https://nij.ojp.gov/topics/articles/best-practices-improving-use-criminal-justice-risk-assessments.

[13] Public Safety Canada, “Organized Crime Research Highlights Number 1,” Accessed September 18, 2025, https://www.publicsafety.gc.ca/cnt/rsrcs/pblctns/rgnzd-crm-rsrch-hghlghts-01/index-en.aspx.

[14] G. Ubertis, “Expert Opinion, Scientific Evidence, and Artificial Intelligence in Criminal Proceedings,” Sistema Penale, 2024, 14–15.

[15] “Partial Ban on ‘Predictive’ Policing and Crime Prediction Systems Included in Final EU AI Act,” Fair Trials, December 11, 2023. https://www.fairtrials.org/articles/news/partial-ban-on-predictive-policing-included-in-final-eu-ai-act/.

[16] “EU AI Act Regulations,” Nemko Digital, 2023, https://digital.nemko.com/regulations/eu-ai-act.; European Union Artificial Intelligence Act: A Guide to Regulatory Compliance, Bird & Bird, 2025, https://www.twobirds.com/-/media/new-website-content/pdfs/capabilities/artificial-intelligence/european-union-artificial-intelligence-act-guide.pdf.

[17] Tetevik Davtyan, “The US Approach to AI Regulation: Federal Laws, Policies, and Industry Self-Regulation,” SSRN, October 17, 2024, https://papers.ssrn.com/sol3/Delivery.cfm/4954290.pdf?abstractid=4954290&mirid=1.

[18] State Council of China, “New Generation Artificial Intelligence Development Plan,” State Council Document No. 35, 2017, Translation available at FLIA (Foundation for a Livable AI).

[19] Filippo Lancieri Laura Edelson and Stefan Bechtold, “AI Regulation: The Politics of Fragmentation and Regulatory Capture,” Oxford Business Law Blog, June 19, 2025. https://blogs.law.ox.ac.uk/oblb/blog-post/2025/06/ai-regulation-politics-fragmentation-and-regulatory-capture.; Peter Cihon, Matthijs M. Maas, and Luke Kemp, “Fragmentation and the Future: Investigating Architectures for International AI Governance.” Global Policy (Wiley Online Library), November 30, 2020, https://onlinelibrary.wiley.com/doi/10.1111/1758-5899.12890.

[20] Eurojust, “7.2 EncroChat: Dismantling of an Encrypted Network Used by Criminal Groups,” In Eurojust Annual Report 2020. Accessed September 18, 2025, https://www.eurojust.europa.eu/ar2020/7-casework-crime-type/72-encrochat-dismantling-encrypted-network-used-criminal-groups.

Europol, “Operation Emma – Dismantling EncroChat, an Encrypted Criminal Network,” Accessed September 18, 2025, https://www.europol.europa.eu/operations-services-and-innovation/operations/operation-emma.

[21] European Commission / CORDIS, “New Tools to Prevent and Fight Organised Crime and Terrorism,” June 26, 2019, https://cordis.europa.eu/article/id/125475-new-tools-to-prevent-and-fight-organised-crime-and-terrorism.;

“PROTON: Modelling the Processes Leading to Organised Crime and Terrorist Networks,” Transcrime (Università Cattolica del Sacro Cuore), March 11, 2024. https://www.transcrime.it/en/publications/proton-modelling-the-processes-leading-to-organised-crime-and-terrorist-networks/.

[22] INTERPOL, “INTERPOL Cooperation Against ’Ndrangheta (I-CAN),” Updated September 10, 2024. https://www.interpol.int/Crimes/Organized-crime/Projects/INTERPOL-cooperation-against-Ndrangheta-I-CAN-Phase-2.; INTERPOL, “INTERPOL-led coalition arrests 100 mafia suspects,” November 6, 2024, https://www.interpol.int/News-and-Events/News/2024/INTERPOL-led-coalition-arrests-100-mafia-suspects.