"Predictive Policing" vs. Presumption of Innocence: The Evidentiary Value of AI Analysis in Tax Evasion Cases

Abstract

This article examines the tension between the deployment of artificial intelligence (AI) systems by ANAF in the fight against tax evasion and the safeguarding of fundamental procedural rights, particularly the presumption of innocence. It underscores the risks posed by "black box" algorithms, which have the potential to shift criminal proceedings from a framework grounded in evidentiary legality to one dominated by probabilistic assessments. The analysis contends that automatically generated reports possess, at most, the evidentiary weight of administrative complaints (extra causam) and cannot serve as independent proof sufficient to substantiate a criminal accusation. Moreover, in light of the reforms introduced by Law No. 126/2024, the study emphasizes the indispensable role of human expertise as a necessary legality filter, ensuring the protection of the right to a fair trial and mitigating the significant information asymmetry between the state and the taxpayer.