Artificial Intelligence in Tax Control: Problems and Prospects of Legal Regulation

Andrianova Natalia Gennadievna PhD in Law Researcher at the Department of Administrative Law and Administrative Process of the Institute of State and Law of the Russian Academy of Sciences 119019, Russia, Moscow, Znamenka str., 10 natalia.g.andrianova@gmail.com Other publications by this author

The issues of using artificial intelligence are currently of great value and are being studied by various scientists, in particular, E.V. Talapina, I.S. Shitkina, however, the issues of legal personality of artificial intelligence ^[1] and the use of this technology in various highly specialized fields, for example, electronic document management ^[2], fall into the circle of issues studied by most scientists. in justice ^[3].

In carrying out this research, general scientific methods (the dialectical method of scientific cognition, the systematic method, methods of analysis, synthesis, generalization, induction, deduction, observation, description of concepts and terms) and special legal methods (in particular, the formal legal method) were used.

First of all, I would like to note that for the Russian Federation, the development of artificial intelligence technology is in accordance with Decree of the President of the Russian Federation No. 145 dated February 28, 2024 [Decree of the President of the Russian Federation No. 145 dated February 28, 2024 "On the Strategy of Scientific and Technological Development of the Russian Federation" // Collection of Legislation of the Russian Federation, 03/04/2024, No. 10, art. It is one of the priority areas of scientific and technological development. In this regard, in the context of the active introduction of artificial intelligence technology, it is necessary to study the content of the concept of "artificial intelligence".

It is interesting to note that the author of the very term "Artificial Intelligence" is John McCarthy, the founder of functional programming [4]. In the regulatory legal acts of the Russian Federation, artificial intelligence is understood as "a set of technological solutions that allow simulating human cognitive functions and obtaining results comparable to at least the results of human intellectual activity when performing specific tasks" [Federal Law No. 123-FZ of 04/24/2020 "On conducting an experiment to establish special Regulation in order to create the necessary conditions for the development and implementation of artificial intelligence technologies in the subject of the Russian Federation - the federal city of Moscow and amendments to Articles 6 and 10 of the Federal Law "On Personal Data" // Collection of Legislation of the Russian Federation, 04/27/2020, No. 17, art. 2701; Decree of the President of the Russian Federation dated 10.10.2019 No. 490 "On the development of artificial intelligence in the Russian Federation Russian Federation" (together with the "National Strategy for the development of artificial Intelligence for the period up to 2030")// Collection of Legislation of the Russian Federation, 14.10.2019, No. 41, art. 5700]. Artificial intelligence technologies are recognized as technologies that include computer vision, natural language processing, speech recognition and synthesis, intelligent decision support, and promising artificial intelligence methods.

In the scientific literature, artificial intelligence is defined as "the result of human intellectual activity, expressed in giving the ability to human-created objects of the material world to perform functions and activities characteristic of human intelligence" ^[5], "a human-created intellectual mind capable of autonomous decision-making in various spheres of society" ^[6]. These definitions of the term "artificial intelligence" emphasize the role of humans in creating this technology, capable of independently making decisions in various spheres of life.

There are also definitions of artificial intelligence in the scientific literature, in which the emphasis is on the artificial nature of this technology, combining an automated mechanism for analyzing and processing information. In this approach, the authors define artificial intelligence as "a property of artificial systems with a knowledge base, a decision output mechanism, and an intelligent interface capable of performing creative functions traditionally considered human prerogative" ^[5], "the process of automated and functional solutions for collecting analytical data, searching for optimal ways to process large amounts of information arrays." and training" ^[8].

Some authors ^[9] separately identify and analyze the threats that may arise when using artificial intelligence technology. Among such threats, there are:

1) threats related to the imperfection of artificial intelligence systems;

2) threats arising from the misuse of such systems.

The first group of threats includes: errors in model training, lack of transparency in decision-making, the likelihood of acting in one's own interests, the possibility of information distortion, weak protection mechanisms, lack of control over development by the developer, the possibility of discrimination, lack of responsibility for the use of artificial intelligence systems, the disappearance of certain professions, job cuts.

The second group of threats arising from the possibility of misuse of artificial intelligence systems includes the following threats: ensuring data security, solving various tasks with the likelihood of obtaining a result that will contain illegal content, the possibility of computer attacks, information collection, which may result in confidential information, information space noise, automation of operations. Nevertheless, with all the existing threats in the field of artificial intelligence, we must not forget about the positive aspects of using this technology.

As some authors note, when developing tax relations in the context of digitalization, the introduction of artificial intelligence technology can occur in two main directions: 1) in the field of tax administration; 2) in the field of changing the rules of direct and indirect taxation of persons using artificial intelligence technologies ^[10].

Of course, a promising area of application of artificial intelligence is its use, primarily in the field of tax control. The Russian tax authorities are recognized leaders in the use of artificial intelligence technology ^[11]. In 2020, the Federal Tax Service of Russia completed the transfer of all taxpayers to the federal database of the automated information system of the Federal Tax Service of Russia AIS "Tax-3", which has many constantly updated and improved functions and subsystems. AIS Tax-3 receives, processes, provides data and analyzes information, generates information resources for tax authorities, statistical data, and information necessary to support management decision-making. The use of this information system allows for the collection and analysis of arrays of data on taxpayers, obtaining information and correlating it with specific taxpayers, tracking the movement of commodity and cash flows, prices applied, and interactions of individuals. The regulatory and legal regulation of the use of this system is carried out by Order of the Federal Tax Service of Russia dated March 14, 2016 No. MMV-7-12/134@ [Order of the Federal Tax Service of Russia dated 03/14/2016 No. MMV-7-12/134@ "On Approval of the Regulations on the Automated Information System of the Federal Tax Service (AIS "Tax-3")"]. At the same time, a detailed analysis of this order allows us to conclude that it does not contain information about how the Tax-3 AIS system directly functions and what technologies are used in its operation. In this regard, some experts highlight "the tendency for public decisions made on the basis of ... data contained in the digital space of the Federal Tax Service, generalized and analyzed by artificial intelligence, over decisions made on the basis of independent analytical work by Federal Tax Service specialists" ^[12].

In addition to the automated information system, in order to improve the quality of desk tax audits of value-added tax returns in accordance with Order of the Federal Tax Service of Russia dated February 10, 2017 No. MMV-7-15/176@ [Order of the Federal Tax Service of Russia dated February 10, 2017 No. MMV-7-15/176@ "On the commissioning of software that implements automation of cross-checks on the functions of desk tax audit of VAT tax returns based on information from purchase books, sales books and accounting logs of invoices issued and received"] software has been put into commercial operation that implements automation of cross-checks on the functions of desk tax audit of VAT tax returns based on information from purchase books, books sales and accounting logs of invoices issued and received, as part of the components of AIS "Tax-3". This software has been named the risk management system "ASK VAT-2". Using this program allows you to automatically divide VAT payers into three risk groups.: high, medium, low. In accordance with the letter of the Federal Tax Service of Russia dated 06/03/2016 № ED-4-15/9933@ [Letter of the Federal Tax Service of Russia dated 06/03/2016 № ED-4-15/9933@ "On the assessment of risk signs of RMS"], the identified risk group is taken into account by the tax authorities when planning and conducting control measures. In fact, the information from the SUR ASK VAT-2 is a set of indicators of a number of sections of VAT tax returns shown in conjunction with similar indicators of VAT tax returns of the taxpayer's counterparties. In accordance with the letter of the Federal Tax Service of Russia dated December 30, 2015 No. ED-4-15/23207@ [Letter of the Federal Tax Service of Russia dated December 30, 2015 No. ED-4-15/23207@// Accounting Appendix (appendix to "Economics and Life"), 2016, No. 6] the use of the SUR ASK VAT-2 system in in-house tax control of VAT declarations allows you to identify tax gaps in an automated mode, in relation to which a set of control measures is carried out. The "gap" type of discrepancy occurs when the SUR ASK VAT-2 program does not detect information about a specific transaction from one of the counterparties of the taxpayer being checked, that is, one of the parties to the transaction has this transaction reflected in the VAT declaration, while the other does not.

In this regard, it is of particular interest to conduct an analysis of judicial practice to assess the possibility of using information obtained using the SUR ASK VAT-2 system to confirm facts indicating that a taxpayer has committed a tax offense.

In the decision of the Tenth Arbitration Court of Appeal dated July 25, 2017 No. 10AP-7846/2017 in case No. A41-74347/16 [Decision of the Tenth Arbitration Court of Appeal dated July 25, 2017 No. 10AP-7846/2017 in case No. A41-74347/16], it is noted that the main purpose of the "ASK VAT 2" system is to check the mirroring of information about reflected accounts-invoices in the buyer's and supplier's tax returns. Automatic online reconciliation of purchase (sales) books is designed to identify discrepancies in the context of each invoice by conducting a virtual counter check of the taxpayer and his counterparties. The information resource "ASK VAT 2" is an internal (non-public) information resource of tax authorities created to automate and systematize the processes of collecting, accumulating, storing and processing certain information about organizations legally obtained by the tax authority in the course of performing its functions. In this regard, the mere presence of discrepancies between the data of the taxpayer's declaration and the data of the tax declaration of his counterparty is not a sign of a tax offense and does not indicate the fact of delivery, which by virtue of the law and under the terms of the contract concluded by the parties must be confirmed by other appropriate written evidence. The court separately noted that the circumstances given in the tax authority's response, based on the use of internal database data, contain only background information. Thus, this court decision emphasizes that the data obtained from the SUR ASK VAT-2 is background information that by itself cannot indicate whether or not a taxpayer has committed a tax offense.

The decision of the Arbitration Court of the West Siberian District dated September 29, 2022 No. F04-5223/2022 in case No. A45-9624/2021 [Decision of the Arbitration Court of the West Siberian District dated 09/29/2022 No. F04-5223/2022 in case No. A45-9624/2021] states that the arguments of the Company with references to the fact that the evidence of the tax authority received through the use of AIS "Tax-3", the PC ASK VAT-2, are inappropriate, were lawfully rejected by the courts, since these information bases represent a single information system of the Federal Tax Service, which provides automation of the activities of tax authorities in all functions defined by the Regulations on the Federal Tax Service, approved by the decree of the Government of the Russian Federation dated 30.09.2004. No. 506. It is noted that the conclusions of the tax authority on the illegality of the VAT deductions claimed by the Company, contrary to the arguments of the taxpayer, are based not only on the data contained in these information resources, but also on the totality and interrelation of other evidence (primary documentation, tax return data, books of purchases and sales, settlement accounts, witness statements, etc.). Recognizing that it was legitimate to hold a taxpayer accountable in accordance with paragraph 3 of Article 122 of the Tax Code of the Russian Federation, the courts reasonably pointed out that the audit established a set of facts indicating deliberate actions to reflect transactions in tax reporting that were not performed and could not be performed by counterparties in order to minimize VAT tax obligations by overstating deductions for disputed transactions. Thus, the consistency of the actions of the participants in the disputed transactions was revealed; the disproportionality of the volume of purchased inventory to actual needs in the absence of the possibility of their supply; the absence of transfers by the taxpayer of funds to counterparties as payment for the supply of goods (performance of contract work); the fact of creating a fictitious document flow that formally meets the requirements of legislation; the complexity and complexity of repetitive (ongoing) actions of the taxpayer that are characteristic of "tax schemes" rather than ordinary activities; a pattern in the actions of counterparties aimed at making it difficult to carry out tax control measures. Thus, in this court decision, a set of other factual circumstances was established, indicating that the taxpayer had committed a tax offense. The data obtained from the tax authority's information system was evaluated as one of a number of other evidence in the case.

The decision of the Arbitration Court of the Moscow District dated May 31, 2021 No. F05-10916/2021 in case No. A40-62388/2020 [Decision of the Arbitration Court of the Moscow District dated 05/31/2021 No. F05-10916/2021 in case No. A40-62388/2020] states that after conducting an audit, the tax authority cannot impute to the most solvent taxpayer in the chain of counterparties any "tax gaps" identified in the "tree of relationships" by the software and hardware complex of the VAT Accounting system for its counterparties of the second, third and subsequent links, without reference to the actual circumstances of transactions, periods and volumes of commodity flows. The court separately noted that the decision of the tax authority does not contain an analysis of the "tax gaps" for VAT, indicating the specific amounts of "gaps", that is, VAT for which the budget has not formed an economic source for its reimbursement, as well as specific companies where such "gaps" arise, and in particular links to which product flows. The judicial act noted that when analyzing the issue of the availability of an economic source for VAT refunds from the budget, the tax authority limited itself to pointing out that "the analysis of counterparties in the information resource ASK VAT 2 does not allow determining the ultimate beneficiaries, since the maximum number of levels of the "tree of connections" is reached. According to the court, this approval by the tax authorities cannot be a sufficient basis for excluding the right to deduct VAT from a taxpayer.

In the decision of the Arbitration Court of the Moscow District dated July 19, 2022, No. F05-9037/2022 in case No. A41-14498/2021 [Decision of the Arbitration Court of the Moscow District dated July 19, 2022, No. F05-9037/2022 in case No. A41-14498/2021], it is noted that the information identified using the ASK VAT-2 software package itself They cannot be evidence of the unlawful presentation of VAT for deduction, but only contain sources of evidence. Their collection, analysis, and evaluation, combined with other circumstances, constitute an independent intellectual activity. The tax authority should not have denied all VAT deductions for the first link, but should have investigated each business transaction at subsequent links. Only by establishing all the facts about the counterparties of the subsequent links could the tax authority come to the conclusion that there was or was not a taxpayer's obligation to be responsible for the actions of all subsequent links.

Thus, in judicial practice, there is a clear approach according to which information systems used by tax authorities are an internal information resource of tax authorities created to automate and systematize the processes of collecting, accumulating, storing and processing certain information about organizations obtained by the tax authority legally in the course of performing its functions. By itself, the detection of "gaps" by the system is not a sign of a tax offense until it is proven by other factual circumstances.

The use of artificial intelligence technology is also possible when conducting tax monitoring as one of the forms of tax control, involving expanded information interaction between the taxpayer and the tax authority in order to ensure the possibility of resolving disputes related to taxation issues in real time, provided that the tax authorities are provided with access to documents for their prompt ongoing verification. In the future, software and technological solutions based on artificial intelligence will allow tax control to cover the entire volume of taxpayer transactions, evaluate them using various criteria, and correctly generate information about the taxpayer's tax liability. Already, the interaction between the tax authority and the taxpayer in the "digital" field is rapidly increasing. Thus, as part of tax monitoring, in accordance with the letter of the Federal Tax Service of Russia dated August 02, 2023 No. SD-19-23/233@ [Letter of the Federal Tax Service of Russia dated 08/02/2023 No. SD-19-23/233@], a period has been set until January 1, 2026, during which organizations can prepare their information systems for integration with AIS "Tax-3". When conducting tax monitoring for periods prior to the specified date, organizations can submit documents to the tax authority either by providing access to their information system or by using the TCS.

Thus, the digital transformation of tax control based on the use of artificial intelligence technology makes it possible to automate tax control processes, however, it seems that in order to ensure a balance of public and private interests, it is necessary to improve legislation on taxes and fees in terms of legislating the use of artificial intelligence technology in tax control with the introduction of new guarantees and new rights for taxpayers.