Applications to Brazilian Criminal Forensics: A Review: UMA REVISÃO

Abstract

Brazilian criminal forensics faces structural deficits — shortage of technological resources, lack of standardization in forensic reports, and team overload — that undermine the quality of forensic investigations, especially in complex violent death cases such as suicides. Artificial Intelligence (AI) and its subfields — machine learning, generative adversarial networks (GANs), diffusion models, retrieval-augmented generation (RAG), AI agents, small language models (SLMs), and native multimodal AI — offer concrete technical mechanisms to overcome part of these barriers. Through an exploratory-descriptive narrative bibliographic review, this article analyzes the technological landscape of Brazilian criminal forensics, maps existing gaps compared to developed countries, and examines, with technical rigor, the scientific evidence on the transformative potential of AI in this field. The bibliographic selection covered approximately fifty sources from Google Scholar, Scielo, and the CAPES Periodicals Portal, published between 2012 and 2025, subjected to methodological quality criteria that prioritized peer-reviewed journals rated Qualis A1–B2. The results show that while countries such as the United States, United Kingdom, and Germany already use AI systematically, Brazil remains at an incipient stage. Technical, financial, training, and regulatory barriers were identified and analyzed. The central proposal of this article is the adoption of a sovereign AI ecosystem — based on locally executed SLMs and national large language models in Portuguese — as a strategy that reconciles technological modernization with the data sovereignty requirements imposed by the LGPD and the digital chain of custody required by the Anti-Crime Package (Law No. 13,964/2019). The article concludes with a concrete five-axis research agenda with specific technical recommendations for transforming Brazilian forensics from a reactive model to a data-driven model supported by what we term the forensic expert's 'cognitive exoskeleton'.