Answers are grounded in this paper and cite their sources.
The article is devoted to the conditions under which results obtained using artificial intelligence (AI) systems can have evidentiary value in forensic examination. It argues that using the single term “AI” in forensic and legal practice is an analytical error, since it groups together technologies with opposite properties–from deterministic expert systems to stochastic generative models.To address the problem, the factual architecture is proposed consisting of three complementary tools: a five-layer model of the evidence life cycle (technical-analytical, expert-interpretive, procedurallegal, organizational layers, and a meta-level); a six-category taxonomy of AI technologies ordered by risk level; and eight criteria of evidence-based (determinism, explainability, validability, function, static character, data provenance, adversarial robustness, and human oversight). Drawing on two matrices –”features–technologies” and “layers–technologies” – it is demonstrated that risk is unevenly distributed and that the key vulnerability of neural-network methods lies not in their mathematical apparatus but in the version control and organizational discipline. As a relevant example, the European Network of Forensic Science Institutes (ENFSI) position is considered. The article concludes with the need to shift regulation from the names of technologies to their properties.
No takes yet. Share an insight, caveat, or question.
Усов et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: