Artificial intelligence, use with caution and limits for tax judges

Artificial Intelligence as "relevant technical support" but never substitutive of judicial activity. The Presidential Council of Tax Justice has deliberated the 'recommendations' for the entry of AI in the tax process, a document that is the substantial synthesis between Maiello's proposal, a declination of last December's SCC resolution, and Cosimo Ferri's.

Net of the numerous references to the EU and national regulatory framework, the resolution emphasises the peculiarity of tax litigation, which due to its "high seriality and rationalisation of information flows" lends itself to a principle of automation from Ai, while at the same time raising "systemic issues regarding the transparency of algorithmic processes, the management of risks of bias and misalignment, as well as the prevention of forms of indirect discrimination and the protection of fundamental rights". Hence the resulting ethical, legal and social risks 'such as algorithmic discrimination, threats to privacy and the question of accountability for autonomous decisions'.

Starting precisely from law 132/2025, according to which the use of artificial intelligence systems in the administration of justice is permitted exclusively as support to the activity of the magistrate, 'excluding any form of automated decision-making subrogation' and firm on the principle that all substantive decisions (the interpretation and application of the law the assessment of facts and evidence; the adoption of final measures), the Council of President Carolina Lussana indicates the precautions to be taken in courtrooms.

The 'operational precautions' include the sovereignty of data and information (never accessible to unauthorised third parties); data protection (never enter sensitive data in the IA, and also consider the risk of re-identification of data); again, data quality (adequate standards in terms of fairness); supervision, i.e. every use of the AI must be supervised by the user to verify compliance with fundamental human rights, data processing, copyright and security regulations, as well as to verify correctness and reliability of the output; human-surveillance (the judge and magistrate will have to correct any unreliable results, reinterpret them or modify them. It must always be verified that the conclusions provided by the IA can be independently replicated). Finally, the individual responsibility: the judge and the tax magistrate is obliged to consciously use the Ai tools in compliance with the regulations, starting from the obligation of information and participation in the training that will be provided on the subject.