Court of Cassation: No to new evidence based on artificial intelligence

New evidence cannot be considered to be that produced using IT tools based on artificial intelligence. The Court of Cassation, in its judgement 29015/2026, rules out the possibility of reviewing judgements on the basis of previously unpublished evidence that is the result of AI. The Supreme Court has ruled that an appeal against the Court of Appeal’s decision to refuse a review of a final conviction – which had been based primarily on wiretaps, for ideological forgery and fraud, committed to gain access to structural and EU funds earmarked for the Lombardy region for tourism purposes. The fraud was carried out by means of an illegal over-invoicing scheme, for the benefit of a company in which the defendant – as had been discovered in the telephone conversation that came to the attention of the investigators – was a hidden partner.

However, with the new evidence provided by the defence, it would have been possible to prove that the person on the phone was not the appellant. The trump card lay in the new method for voice analysis used by the experts, which involved ‘comparison software utilising artificial intelligence, or based on the comparative analysis of biometric data’. The Court of Cassation acknowledges the possibility of submitting new evidence, based on more advanced and sophisticated technologies, which provide new objective data on which to base different technical assessments. In the case under consideration, however, the appellant had not provided any clarification regarding the ‘new frontier’ of the system employed, on the basis of which the judges would have had to overturn the unfavourable verdict, thereby returning to the conclusion reached by the judges at first instance who had acquitted the defendant.

The guidance provided was vague and lacked any real substance regarding the methodologies actually used, their scientific reliability and their genuinely innovative nature, in relation to the comparison of biometric data.

The only certainty is that the new tests made use of an AI-based programme designed to identify a person by their voice. The Supreme Court does not consider this sufficient.