Balancing Innovation and Oversight: Tax Advisers Critical for AI

Introduction

The EU aspires to create an artificial-intelligence-friendly ecosystem where economic players can find infrastructure, research facilities, financial means, legal frameworks, and people with adequate skillsets to invest in and deploy AI. Those ambitions were clarified by the Coordinated Plan on AI released in 2018, which delineated the bloc’s aim to render the EU a global leader in developing and using AI, promoting a human-centric approach and ethics-by-design principles, as well as integrating AI (among other systems) into public administration.¹ In 2020 the White Paper on AI² was released, proposing a framework for regulating high-risk AI applications and initiating public consultation. Finally, in August the AI Act³ came into force, making the EU the first regulatory landscape with comprehensive AI regulation. The act’s primary goal is to ensure that AI systems developed and deployed within the EU are safe, ethical, and trustworthy.

Similarly, the OECD’s vision for tax administration is digital transformation, which is reflected in a landmark report published in 2020.⁴ The report called upon tax administrations to establish a common language and framework to assist countries in their digital transformation journey. According to that report, approximately 50 percent of tax administrations around the globe are using AI systems for risk assessment and fraud detection.⁵

This article outlines some of the ways AI can be applied in the tax world. Tax authorities globally are increasingly leveraging AI technology to enhance their capabilities, from automating audits to improving fraud detection to ensuring greater compliance. In this sense, I hope the present article raises awareness among taxpayers of the growing sophistication of tax authorities’ tools and shed light on the continued importance of skilled advisers when navigating complex regulations and ensuring accuracy in a rapidly evolving technological landscape.

Tax Compliance and Strategy

AI may enhance tax compliance experience through automated systems. AI tools may assist taxpayers through automated preparation and submission of their tax returns and by providing personalized tax consulting services. The use of generative AI (like ChatGPT) has already provided us with a taste of the possibilities of AI systems. How far are we from a reality in which a taxpayer can receive personalized guidance regarding their tax position from a virtual tax adviser? Will we ever get to the point at which submitting tax returns becomes unnecessary because they will be prepared and submitted automatically by secure tax technology?

The Spanish tax administration has developed a virtual assistant tool for VAT using AI with the purpose of advancing toward quality information provision with unique criteria and facilitating the understanding of complex regulations. The system provides information on registering and correcting invoices, obligations related to foreign trade, chargeability, taxable amounts, tax rates, exemptions, and deductions on real estate transactions via a chatbot available to both taxpayers and tax officers. The virtual assistant helps users find the information they need to get the answers they are looking for.⁶

The Greek tax administration provides an AI assistance system to help taxpayers prepare and submit tax returns and has intelligent virtual assistants (smart agents) who can provide “quasi” tax advisory services directly to taxpayers.⁷ However, while automation can streamline processes, it may also introduce biases in interpreting tax laws or take technical positions regarding tax return preparation. Automated systems may oversimplify nuanced legal requirements or fail to consider unique circumstances that could affect compliance, making it crucial for taxpayers to seek expert human advice.

Another recently developed AI application may predict judicial decisions in tax matters, for which the term “predictive justice” has been coined. Predictive justice includes data analytics functions, machine learning functions, and techniques capable of analyzing large volumes of court decisions to make predictions about the outcomes of tax cases. A recent example is the Blue J project, developed by a private company in Canada, cofounded by professors at the University of Toronto, and provides taxpayers with answers to tax questions that have been addressed by the Canadian tax courts.⁸ It can classify taxpayers as self-employed or salaried for income tax purposes by looking at how the courts have interpreted the law in the past. The system is also designed to answer other types of questions, like whether a person should be considered a resident for tax purposes, whether expenses should be considered deductible, and so forth. However, University of Toronto Law professor and co-founder of Blue J Abdi Aidid talked about the limitations of AI predictive justice at the ABA conference, mentioning that “the really important thing when you’re using a tool like [ChatGPT] is recognizing its limitations.” He further explained that the tool “is not providing source material for legal or tax advice. What it’s doing — and this is very important — is simply making a probabilistic determination about the next likely word.”⁹ As Aidid has highlighted, AI technology requires professional oversight to ensure that their outputs are accurate and aligned with the nuanced interpretations of tax laws. In an era of growing tax sophistication, involving tax advisers is more essential than ever to ensure that tax strategies are both compliant and optimized for the specific needs of individuals and businesses.

Fighting Tax Evasion, Smuggling, and Fraud

AI includes techniques that may be used to analyze large amounts of data and extract relevant information (data mining). Tax administrations use data mining to enhance the collection, organization, and processing of tax data in a systematic and transparent manner.

In September 2023 the IRS announced that, in an effort to restore fairness in tax compliance by shifting more attention onto high-income earners, partnerships, and large corporations, it would use improved technology as well as AI.¹⁰ According to the IRS’s strategic operating plan for fiscal years 2023 through 2031, the agency will use data and technology to ensure that resources are focused on noncompliant taxpayers.¹¹ The key objectives are delivering cutting-edge technology, data, and analytics.

Following the same direction, in December 2023 Greek tax authorities announced their plan to incorporate AI in the fight against tax evasion and tax avoidance.¹² As AI systems may collect useful information from many datasets from various sources (like banking transactions, digital public platforms, and social media platforms), AI may detect in real time suspicious movements (which could indicate tax fraud, evasion, or avoidance), unusual discrepancies in the income of individuals and businesses, and unusual transactions or money transfers. Typically, tax audits start after the end of the tax year in question (ex post approach), while the AI attributes an element of “continuity” to the tax audit. Additionally, the taxpayer’s rights should always be respected based on the EU Charter of Fundamental Rights. Regarding tax audits driven by AI technology that continuously collects data from various sources, the emphasis should be on the right of the taxpayer to a defense and equality of arms (based on article 47 of the EU Charter of Fundamental Rights), which means that the taxpayer should have access to the file and all relevant information and documents that the authorities use for the tax audit.¹³

AI is expected to play a significant role in the field of transfer pricing, as well. By using AI-driven analytics, tax authorities could process and analyze large datasets related to intercompany transactions, thereby identifying discrepancies and patterns that may indicate transfer pricing manipulation. For instance, machine learning algorithms could benchmark pricing against similar transactions within the same industry, highlighting outliers that warrant further investigation. AI tools could quickly gather relevant data and assess compliance with local and international transfer pricing regulations. This would not only accelerate the audit process but also enhance the accuracy of assessments, allowing tax authorities to address potential tax base erosion and ensure compliance with the arm’s-length principle.

As data collection and processing from various sources increases, the need for transparency and confidentiality of taxpayer information becomes crucial. For example, the French Constitutional Council oversaw the constitutional review of certain provisions introducing a temporary (three-year) instrument that would grant tax authorities the power to gather taxpayer data with automatic AI tools. These tools would allow for the collection of information that is publicly available on the internet (including social networks and websites to sell goods and services), which the tax administration could use for investigating tax and customs infringements.¹⁴ The French court, considering the proportionality principle, concluded that the automated AI mechanism would restrict taxpayers’ fundamental rights, so it set limits on its use in tax proceedings.¹⁵

Also, regarding the use of AI in the course of tax audits, it is important to observe that AI functions should be evaluated in light of article 22 of Regulation (EU) 2016/679, which excludes the automated use of algorithms for the purposes of administrative decision-making.¹⁶ To apply the prohibition established by article 22(1), four conditions are needed: (1) A decision must be made, (2) it must be solely based on automated processing, (3) it must include profiling, (4) it must have a legal or significant effect.¹⁷

Finally, AI may be relevant for administrative cooperation in direct taxation between the competent authorities at the European and international level, which entails (among other things) exchange of information for tax purposes (EU Administrative Cooperation, notably a series of directives that facilitate the exchange of tax information between EU member states¹⁸), tax reporting (for example, the OECD’s common reporting standards, which operate at an international level and facilitate the automatic exchange of information between countries to combat tax evasion¹⁹), and cross-border joint tax audits promoted by the OECD, which involve tax authorities from multiple countries working together to examine the tax affairs of multinational entities.²⁰ AI systems may play a vital role in future joint tax audits because it would allow tax administrations to extract information from large datasets in real time and effectively enhance cross-border cooperation. AI may also assist in comparing interpretation trends of tax concepts (like permanent establishments and the principal purpose test) in different states to further assist administrative cooperation between states.

As AI enhances the efficiency of tax audits, the above limitations make the role of tax experts indispensable. During tax audits, it is important that tax advisers are involved to advocate for taxpayers’ rights, ensure transparency, and facilitate communication between the tax authorities and the audited entity. Tax experts who are knowledgeable about AI could efficiently address misunderstandings that arise from AI assessments or discrepancies between AI-generated scenarios and the facts and circumstances of the case in question. Finally, transfer pricing specialists can assist in interpreting and assessing AI-generated findings.

Tax Law Regulation and Policy

European tax laws often come from long-standing public and EU interservice consultations by different departments of the European Commission, working groups of the council, consultations at the European Parliament level, and review by the relevant parliamentary committee. These processes and consultations take place in English and French. In this respect, some tax and legal concepts that were drafted in English are not easily convey across foreign languages. The same difficulty can be observed during cross-border tax agreement negotiations and at the integration stage of European tax rules into domestic law systems. AI may assist with these difficulties because it may produce technical translations simultaneously in multiple languages ​​in a way that mimics human scientific methods, achieving the most accurate translation of tax and legal concepts in various languages, thus improving the efficiency of technical discussions between states.²¹

In the field of international tax law, multilateral agreements are gaining ground over bilateral cross-border agreements (for example, the Multilateral Convention to Implement Tax Treaty Related Measures to Prevent Baser Erosion and Profit Shifting, with over 100 states participating in consultations). The consultation and negotiation processes for multilateral instruments with numerous members make the role of AI even more crucial for the linguistic effectiveness of said processes.

As new tax policies are shaped, tax experts must always ensure that AI-assisted processes do not overlook the legal protections and entitlements of taxpayers. AI serves as a powerful tool to enhance efficiency, but tax specialists are key to ensuring fairness, accuracy, and the protection of human rights throughout public consultation and legislative processes.

Cross-Border Dispute Resolution

The same linguistic and interpretative challenges may be encountered in mutual agreement procedures, a mechanism that contracting states use to resolve disputes or difficulties caused by inconsistencies in the interpretation and application of double tax treaties. MAP is a public international law procedure, which is independent from domestic law. MAP typically involves consultation between the competent authorities of the contracting states, allowing them to communicate directly to resolve matters through a cooperative process. The exchanges may take place through various channels, including document exchanges, letters, faxes, telephone calls, face-to-face meetings, or any other modern means of communication. One of the core difficulties in MAP, however, is interpretating tax concepts that often vary across jurisdictions. Because these discussions frequently involve tax terminology grounded in domestic law systems, precise translation and clear understanding are crucial. AI can significantly enhance the efficiency of MAP by providing real-time, technically accurate translations that capture legal nuances of different jurisdictions.

Tax advisers and legal professionals play a critical role in reviewing AI-generated outcomes in the course of a MAP. This collaborative approach between AI technology and human expertise fosters clear communication, safeguards integrity in the dispute resolution process, and ultimately protects taxpayers’ rights.

Concluding Remarks

While AI is revolutionizing tax administration by enhancing efficiency and compliance through automation, predictive analytics, and personalized assistance, it is not without limitation. Human oversight remains essential to ensure the accurate interpretation of tax law and to prevent costly errors. The growing sophistication of AI systems in tax enforcement highlights the increasing need for skilled tax advisers who can navigate these tools, provide expert advice, and tailor compliance strategies to meet business-specific needs. In this evolving digital landscape, the collaboration between AI and tax professionals will be essential to ensure that AI’s potential is fully realized while safeguarding fairness and accuracy.