In the recent years, Artificial Intelligence (AI) has taken center stage in most public administrations of the European Union, and tax authorities have also introduced such mechanisms in their processes. Indeed, most Member States have now reported using some sort of AI-powered techniques across a number of their functions. In the tax sphere, these tools are mostly used to ensure taxpayer compliance, through either improving compliance-specific administrative processes, enhancing taxpayer services (for instance, pre-filing systems of tax declarations) or facilitating policy review. Nonetheless, although most Member States are using AI tools, a number of major differences across Europe have been identified in multiple layers, ranging from the extent of their use, the applicable legal framework (or lack thereof), the technology being used, issues of interoperability that may arise and finally, differences in the administration’s ability and resources to deal with the imperatives stemming from the use of such tools. In addition to these disparities, the ‘traditional’ problems that relate to the use of AI in public administrations in general appear also in tax administrations (including biases, lack of transparency, issues of personal data protection).
As taxpayers’ cross-border activities have increased tremendously in the past decades, inevitably, more than one tax authorities are involved in the taxpayers’ tax assessments. This increase, together with the absence of a harmonized legal framework in direct taxation, has led to more opportunities for tax planning, tax avoidance, even tax evasion, challenges that tax administrations aim to curb, inter alia, through the use of AI. This use, however, may impinge upon taxpayers’ rights. In this context, it is important to examine whether taxpayers may seek protection in the right to good administration as enshrined in Article 41 of the European Charter of Fundamental Rights (the Charter).
Transposed in the tax administrations’ framework, this right encompasses a taxpayer’s right to have their affairs handled impartially and fairly by the administration in compliance with the principle of equality, the right to be heard, the right to be given reasons for an administrative decision, as well as the right to have access to one’s own files to facilitate the reviewability of the tax administration’s decision. All these manifestations of the right to good administration contribute (also) to safeguarding the fundamental principles of legality and legal certainty. At the same time, the right to good administration may affect tax administrations that use AI tools, especially in what regards transparency and explainability. This is even more important in taxation as most AI tools are geared towards compliance management and they are, therefore, more likely to lead to adverse decisions for the taxpayer that may, eventually, require review (either administrative or judicial).
As the investigative stage of tax investigations remains largely secret (Berlioz, Othymia) and the taxpayer is not recognized any rights during this stage, and since tax administrations are relying more and more on AI to detect incompliance, taxpayers would benefit tremendously from the protection of Article 41, as they would finally gain access to otherwise unavailable information and procedures (e.g. right to be heard). Despite its obvious value for the taxpayers, several limitations make the application of the right to good administration questionable. The first major obstacle to overcome is the applicability of the Charter itself in the area of direct taxation, a largely unharmonized area. This more general issue becomes all the more problematic as tax administration constitutes the largest part of public administration given that everybody is subject to taxes or is considered in/for someone’s tax assessment. But even if the door to the applicability of the Charter in direct taxation were to be opened through the continuously increasing EU legislation, the unclear scope of the right to good administration (its application to administrative ‘decisions’) together with the inherently difficult nature of determining the role of AI in the administration’s final outcome would not guarantee the application of the right. This is because, in order to rely on Article 41 of the Charter, one would first have to understand whether the act at issue constitutes a decision as per Article 41 and second, to determine how much the AI tool has contributed to this administrative decision. In this context, for example, if the tax administration were to use an AI tool to select a taxpayer for an audit, two points would deserve further clarification: a) what and when is the administrative decision made: when the taxpayer is notified of the audit or when the tax administration decides to proceed to a tax adjustment on the basis of the audit; b) what has been the role of the AI in that decision. This last question is also fundamental, in that the use of AI may affect the (specifics) of the protection granted by Article 41 (including issues of explainability and human oversight).
A second scope-related limitation concerns the right’s scope ratione personae, as Article 41 Charter is addressed to any person dealing with the institutions, bodies, offices and agencies in the Union. This is quite restrictive, as it means that the right only provides protection to taxpayers that deal with an EU body using AI for tax purposes. This is a very unlikely scenario as no central EU tax administration exists; instead, tax administrations are domestic bodies and, therefore, fall outside of the scope of the article. There is one potential exception to this, the Transaction Network Analysis (TNA) tool used by the Eurofisc cooperation mechanism, which uses AI in order to facilitate the detection of cross-border VAT frauds through joint action between Member States and the EU. In this case, one could argue that the use of AI by the TNA could fall within the scope of the right, although it is likely that Article 41 would still not apply as the final decision stemming from the use of the AI is most likely going to emanate from a domestic administration rather than an EU body.
Apart from the TNA, all AI tools are used domestically, and, thus, Article 41 cannot be invoked. While such a finding would be disappointing for leaving unprotected (through the Charter) all EU citizens, one solution may be found on the principle of good administration as this has been developed by the CJEU. According to case law (Bezirkshauptmannschaft Hartberg-Fürstenfeld) the principle of good administration provides similar guarantees to the right, but, unlike the right, it applies to domestic administrations as long as they are implementing EU law. Therefore, as long as tax administrations are using AI tools in their fulfilment of EU obligations, taxpayers would be able to rely on this principle (Hernandez). Nevertheless, this opportunity comes with two distinct caveats which may render its use difficult. On the one hand, the protection offered by a jurisprudence-based principle is necessarily harder to rely on than the protection associated with the Charter’s rights, as courts have more leeway in applying the former. On the other, determining when a tax administration is using AI in order to fulfil EU law-imposed requirements may prove to be more complicated than initially thought. This is because, the most relevant piece of legislation in direct taxation, the so-called ‘DAC’, the Directive on Administrative Cooperation in the area of direct taxation, imposes on tax administrations the obligation to collect and (automatically) exchange taxpayers’ information with other Member States. Because of this ‘multi-use’ purpose, it becomes difficult to discern what constitutes ‘implementation of EU law’. If one adds evaluating the role of AI in this implementation (that could potentially open new doors to taxpayers’ protection), things become even more perplexed. This general problem is particularly evident in the field of direct taxation, a largely unharmonized area, outside the EU competence. In contrast, in the area of indirect taxation, taxpayers may rely on the VAT directive to invoke the principle of good administration, according to the CJEU (Cridar).
In the DAC context, one tax administration may proceed, for instance, to a tax adjustment on the basis of an information it received through the use of AI tools. The question that arises in this context is whether this is enough to trigger the application of the principle given that the CJEU has stated explicitly in Hernandez that in order to be implementing EU law, a Member State should find itself within the confines of an obligation set by EU law, and it is not sufficient that the EU legislator regulates one area of the law for the principle to apply to this area in its entirety. Thus, in relation to the DAC if the CJEU case law were to be understood in a broad sense, the taxpayers could rely on the principle of good administration if an AI-based administrative decision ensues from the obligations in the directive. That would be the case, for example, if an AI tool were used to facilitate the exchange of information or to classify reporting information collected. For instance, should the taxpayer be able to rely on the principle if a Member State proceeded to a tax adjustment on the basis of information it received with the use of an AI tool, on the basis of the DAC? One such case would be when an intermediary reports a potentially aggressive tax planning scheme to the tax authority, as required by the so-called DAC 6, and the latter ‘flags’ through the use of AI the taxpayer, leading to a tax adjustment or a penalty. This, however, would work only on very specific cases and it is doubtful that this possibility would extend to any administrative decision taken on the basis of an AI tool using the information collected or exchanged. Beyond tax-specific legislation, the EU recently started regulating the area of technology and intends on doing the same about AI use. Nevertheless, it is doubtful whether the up-and-coming AI Act would make any difference in the applicability of the principle in taxation, as most of the obligations imposed in the Act are limited to ‘high-risk AI’ tools, the definition of which seems to effectively exclude tax administrations from its scope.
Beyond these issues, the applicability of the principle can also be put into question due to the fact that AI use in tax administrations may fall under the procedural autonomy enjoyed by Member States, when implementing EU law. Despite the limitations of this autonomy by the principles of equivalence and effectiveness, the possibility for taxpayers to successfully invoke the principle of good administration appears to be further reduced.
All in all, under the current circumstances the principle of good administration could prove to be a useful tool for taxpayers facing an AI-driven administrative decision in the area of VAT. In the area of direct taxation, however, the application of the principle must be assessed on an ad hoc basis that depends on the specifics of the EU legislation in place and the interpretation of the principle by the CJEU. One may wonder, therefore, whether taxpayers dealing with the use of AI by tax administrations (in direct taxation cases) are completely deprived of the protection afforded by the right (or the principle) of good administration. The answer is that, for the vast majority of cases, yes! Despite this unsettling outcome, it has to be acknowledged that this is fully in line with the letter of the Charter that requires that EU fundamental rights apply when Member States are implementing EU law. This is not only in line with Member States’ procedural autonomy but also with their (alleged) retained sovereignty in direct taxation. Thus, despite the positive effect a potential extension of the EU protection to national tax administrations would have, in that would establish a minimum level playing field consistent with the rule of law desiderata; under the status quo the application and the content of good tax administration is left to the Member States.