Pursuant to Article 21 of the Mexican Tax Administration Service Law, such entity of the Mexican Ministry of Finance and Public Credit is required to draft and publish on an annual basis a continuous improvement program that sets specific tax-related goals, such as the fight against tax evasion and avoidance, an improvement in tax collection or a decrease in collection costs. Thus, for the current year, the Mexican Tax Administration Service (“SAT” due to its Spanish acronym) published the “Master Plan 2024”, whereby it identified vulnerable activities, taxes on foreign trade and domestic taxes, improper applications of balances in favor of value added tax, the fuel market, simulated transactions, among others, as the main priorities to be audited.
Over the years, the SAT has been pursuing ways in which it may follow up on the auditing of transactions that threaten Mexican revenue through its verification authority provided for in Article 42 of the Federal Tax Code, the most aggressive being the domiciliary visit (visita domiciliaria) and the incorporation of the electronic tax review in subsequent years. However, the federal tax authorities have realized that their efforts have not been sufficient, and therefore in the “Master Plan 2024” it refers to the implementation of graph analytics and machine learning models, an artificial intelligence (“AI”) program based on machine learning for the classification of risky operations and identification of taxpayers who engage in evasive and elusive practices, as well as for the detection of inconsistencies in digital tax receipts through the Internet (“CFDIs” due to its Spanish acronym).
Machine learning is a program that the SAT is currently using to be able to collect more efficiently. This technique uses algorithms that detect errors that are built automatically until no more errors arise, which shall be more practical for the efforts of the SAT, given that it was not able to cope with the assigned employees, nor with the infrastructure it uses to follow up in a traditional way on its auditing authority.
However, the simple idea of taxpayers being audited through AI causes discomfort at first glance due to two relevant aspects: (i) the fear that a machine might compensate for the deficiencies of the tax authorities and it is more likely that a greater number of taxpayers might be identified in simulated transactions, and (ii) because it is possible that the fine line between auditing and the violation of human rights such as identity, dignity and security might be breached. In this regard, in 2019 the Organization for Economic Cooperation and Development (“OECD”) issued certain recommendations to governments for the design and use of AI, being the following: (i) AI should be primarily at the service of people; (ii) AI should be designed in a way that respects the rule of law, human rights, democratic values and diversity; (iii) the management of AI should be conducted in a transparent and accountable manner, so that individuals are aware when they are interacting with it and may challenge the outcomes of such interaction; (iv) AI should operate reliably and safely, and potential risks should be assessed and managed at all times; and, (v) the individuals or organizations that develop or manage it should be accountable for its proper operation in relation to the foregoing.
Regarding the foregoing, questions naturally arise about the proper handling of the IA by the SAT, since in many cases obvious abuses committed by such tax authority have been disclosed. However, being objective, it must be recognized that many taxpayers are constantly searching for new ways to avoid paying their taxes, declare less income than obtained or simulate operations that in reality do not occur, for which the use of the AI could be a very fair way to detect such fraudulent transactions, without falling into the extremes of the past, such as the Mexican criminal reform of 2019, which considered the crimes of smuggling, fraud and the acts of Article 113 bis of the Federal Tax Code related to companies that invoice simulated operations (EFOS due to its Spanish acronym) and deduct simulated operations (EDOS due to its Spanish acronym), as acts that threatened Mexican security and were, therefore, considered as organized crime offenses.
Although the SAT has achieved important advances in digitalization, such as the use of the tax mailbox (buzón tributario), the electronic tax review and the requirements for the issuance of CFDIs provided for in Articles 29 and 29-A of the Federal Tax Code, there is still a long way to go for the federal tax authority in terms of increasing collection efficiency, reducing tax evasion and avoidance, combating corruption and improving taxpayer service. Therefore, it is my opinion that, following the principles on AI proposed by the OECD, the use of AI in tax matters is not only the right choice, but also a necessary means for the federal tax authority to be able to fulfill its duties. The AI shall help it to be a means of identifying the income and expenses generated by taxpayers and to have a record of transactions in order to identify in real time the operations that may be simulated.
In conclusion, the incorporation of AI is a necessary mechanism for the SAT to prevent tax evasion and avoidance practices, as long as clear and public management parameters are followed to avoid violating taxpayers’ human rights such as security and privacy, to reduce abusive acts by the federal tax authorities, and to ensure that the results obtained by AI are always transparent and auditable.
Mauricio Iturralde Punzo
