Italian startup Rozes uses machine learning to scan corporate balance sheets for anomalies linked to organized crime, aiding public procurement and AML.

Financial statements are often viewed as dry summaries of revenue, debt, and profit. However, for the founders of Rozes, these documents are more like digital fingerprints. By applying machine learning to corporate accounting, the Italian startup is attempting to uncover the invisible traces left by organized crime within the legal economy.

Decoding the financial DNA of organized crime

The premise is straightforward but technically complex: criminal organizations do not leave a line item labeled 'mafia activity' in their books. Instead, they leave statistical anomalies. These are specific combinations of financial ratios, corporate behaviors, and accounting irregularities that, when viewed in isolation, might seem benign or merely inefficient. When viewed as a cluster, however, they mirror the operational patterns of firms previously convicted of money laundering, fraudulent bankruptcy, or the issuance of fake invoices.

Founded in 2019 and later recognized as a spin-off of the University of Padua, Rozes has transitioned academic research into a commercial risk intelligence platform. The system does not look for a single 'smoking gun' entry. Instead, it analyzes an articulated set of signals. If a company's financial profile closely overlaps with the known patterns of criminal entities, the algorithm assigns it a high risk score.

The data engine behind the algorithm

The efficacy of any AI is dependent on its training set. To build a reliable model, researchers Francesco Ambrosini, Michele Fabrizi, and Antonio Parbonetti from the University of Padua utilized a rigorous dataset. The algorithm was trained using an archive of 2,082 companies that have established judicial links to organized crime.

By feeding the machine learning model the balance sheets of these proven criminal fronts, Rozes created a reference sample of 'criminal accounting.' The AI now compares the data of any target company against this benchmark, identifying similarities that a human auditor might miss across thousands of pages of documentation. This approach transforms due diligence from a manual sampling process into a systemic scan of the entire corporate landscape.

Risk scoring versus judicial certainty

A critical distinction in the Rozes methodology is the difference between detection and accusation. The platform is designed for due diligence, counterparty verification, and anti-money laundering (AML) activities, but it does not issue legal verdicts. It does not declare a company 'mafia-owned'; rather, it indicates where investigators should look more closely.

The algorithm does not establish who is a mafioso. It searches for anomalies and indicates where it might be necessary to look better.

For banks and public administrations, this distinction is vital. The value of the tool lies in its ability to narrow the field of suspicion. When dealing with hundreds of thousands of suppliers or clients, it is physically impossible to conduct a deep-dive investigation into every entity. The AI acts as a filter, prioritizing high-risk targets for human investigators and judicial authorities.

Securing public procurement and PNRR funds

The practical application of this technology is already moving into the public sector. The Italian Ministry of Infrastructure is currently experimenting with the Rozes system to scrutinize tenders related to the PNRR (National Recovery and Resilience Plan). Given the massive influx of European funds, the risk of infiltration by organized crime into public works is a primary concern for the Italian state.

By integrating this AI into the procurement process, the government aims to protect public contracts from fraud and money laundering. The system allows authorities to screen bidders more effectively, ensuring that funds intended for infrastructure and innovation do not inadvertently fuel criminal networks through shell companies or fraudulent consortia.

The evolution of a university spin-off

The journey of Rozes reflects a broader trend of academic research entering the fight against financial crime. In May 2026, the University of Padua formally renewed the startup's spin-off status. This renewal was not merely administrative; it was based on a rigorous evaluation of economic solidity, ongoing research projects, and the strength of the company's patents.

This academic backing ensures that the tool remains grounded in scientific methodology rather than 'black box' AI logic. The collaboration between the university and the startup allows for continuous refinement of the algorithm as criminal organizations evolve their methods to evade detection. As the mafia adapts its financial strategies, the AI can be retrained with new judicial evidence to keep pace.

Global implications for international business

For entrepreneurs and firms operating in the USA, UK, and global markets, the emergence of tools like Rozes signals a shift in the landscape of corporate transparency and KYC (Know Your Customer) requirements. While the current application is focused on the Italian context, the underlying logic of 'pattern-based risk scoring' is universally applicable to any jurisdiction where organized crime infiltrates legal businesses.

In the United States and the United Kingdom, where AML regulations are stringent, the adoption of such AI tools could redefine the standard of 'reasonable due diligence.' Companies that rely on traditional checklists may find themselves lagging behind regulators who use machine learning to spot systemic risks. Furthermore, as the EU continues to refine the AI Act, the use of AI in law enforcement and risk assessment will face strict transparency requirements, making the 'explainable AI' approach—where patterns are linked to judicial evidence—the only viable path forward.

International firms partnering with European entities, particularly in infrastructure or construction, should anticipate a higher level of scrutiny. The ability to scan thousands of balance sheets in seconds means that the 'corporate veil' used by shell companies is becoming increasingly transparent. For the global entrepreneur, this means that the integrity of the entire supply chain is no longer just a compliance checkbox, but a data-driven necessity.

FAQ

---

Sources: Libero, Repubblica, Venetoup ·