Abstract
Artificial intelligence (AI) is entering criminal proceedings faster than forensic tactics governing interrogations, confrontations, on-site verification, and clarification of testimony can be rethought. These verbal investigative actions remain the stage at which the fairness of a criminal process, and hence access to justice and strong institutions, is most exposed to coercion, error and false confession. The study develops a scientifically grounded model of how the tactics of verbal investigative actions are transformed when an investigator gains access to AI tools, and it tests three propositions about the assistive nature of AI, the loop structure of its influence, and the need for legally informed data labelling. The study employs a doctrinal, comparative-legal, and modelling design, applied to 25 scholarly sources published in 2010–2026, five normative acts of the Republic of Kazakhstan and the European Union, the practice of five European jurisdictions, and the technical specification of a patented prototype developed under project BR28712219. The results show that the tactical effect of AI is distributed across four functionally linked loops (preparatory, communicative, recording-analytical and control), that each loop carries one dominant risk and one procedural safeguard, and that the prototype records four synchronized video channels at 1920×1080 and at least 25 fps together with 16 kHz PCM audio, the format required by current speech-recognition engines. The findings of this study suggest that AI transforms verbal investigative actions by changing the quality of information support rather than the subject of decision-making, that a human-centric model is the only one compatible with procedural safeguards, and that responsible innovation in this field depends on a systematic annotation methodology, because arbitrary labeling reproduces annotator errors instead of forensic patterns.