State of use of AI systems
The European Union and the European Commission in particular have always been strong proponents of the digitalization of national tax administrations and the integration of AI systems to tax processes.
In 2005, the EU adopted the Customs Risk Management Framework (CRMF), to facilitate the exchange of taxpayer data for risk-management and profiling.
In 2006, the Commission proposed a tax compliance risk-management (CRM) framework and outlined early iterations of machine-learning systems. For instance, the 2006 CRM Guidelines presented XENON, a webscraping tool used at the time by the Netherlands, Sweden and Denmark to detect unidentified taxpayers on online platforms. Thus, the EU through the Commission and the Fiscalis Network, has adopted an active role to promote the integration of machine-learning to tax compliance processes.
What functions are performed with AI systems?
Since taxation remains mostly a competence of Member States, the range of tools proposed at EU level is limited. Nonetheless, to facilitate the detection of cross-border and international offences, the EU has deployed certain tools at the disposal of domestic authorities:
- Real-time risk detection (TNA): Transaction Network Analysis (TNA) is a real-time data matching system meant to detect and prevent missing-trader and carousel fraud. The system cross-verifies VAT transactional data with data on the platform for cooperation and information exchange between national tax officials. TNA also cross-check the information with criminal records, databases and information held by Europol and the anti-fraud agency (OLAF). The system can flag suspicious transactions and mismatches in real-time, and enable swift liaison between domestic agencies concerned. As reported by the Commission, TNA enabled the recovery of 12.7 billion euros in 2023. TNA was developed by Belgium and extended to other EU Member States. TNA is now available and used by all Member States of the EU.
- Real-time risk detection (CRMS2): Similarly to TNA, the Customs Risk Management System 2 (CRMS2) is first and foremost a platform meant to facilitate the exchange of tax data between domestic customs agencies. Nonetheless, CRMS2 is equipped with a real-time risk detection system that can flag suspicious transaction on a broad range of possible risks. When goods are stopped at an external border for any of the risks covered, e.g. explosives, risks to health, commercial and financial risks, etc. CRMS2 enables instant sharing of the risk (so-called RIF ‘Risk Information Form’) to other customs offices and its inclusion into compliance risk registers.
What data is processed by these systems?
Regarding TNA, the Eurofisc enjoys a broad mandate to access and share taxpayer data. The data processed by TNA includes at least: cross-border transactional information, data on the VAT Information Exchange System (VIES), criminal records, as well as information held by Europol and OLAF.
Regarding CRMS2, domestic customs authorities enjoy an equally broad mandate to access and share customs data. CRMS2 provides a central database which connects all international airports, sea ports and major land borders. Accordingly the risk detection system may potentially process any data of operational officials and risk analysis centres in the Member States of the EU.
Are these systems regulated by specific norms?
Although the modalities on the exchange of information for VAT and customs purposes are subject to specific EU Regulations, respectively Council Regulation 2017/2454 (VAT) and Regulation 952/2013 (Customs), the underlying AI systems are not regulated by ad hoc norms.
