Artificial Intelligence Risk Classification: A Comparative Analysis between European Regulation and the Brazilian Judiciary

UMA ANÁLISE COMPARATIVA ENTRE A REGULAÇÃO EUROPÉIA E O JUDICIÁRIO BRASILEIRO

Abstract

The rapid evolution of artificial intelligence demands robust regulatory frameworks that address potential risks to procedural integrity, impartiality, and the protection of fundamental rights. The European Union’s Artificial Intelligence Act (AIA) categorizes AI systems into different risk levels. However, this static reliance on application fields may result in inaccurate estimations of risk magnitude, particularly with regard to general-purpose systems that exhibit unpredictable applications.

The core problem of this investigation lies in identifying the gaps and challenges within current AI risk classification models, both in European regulation and the Brazilian judiciary, and proposing approaches that enable a more precise and dynamic assessment of such risks.

The underlying hypothesis is that integrating more sophisticated risk assessment frameworks—such as the one developed by the Intergovernmental Panel on Climate Change (IPCC)—can enhance the predictive and risk management capacities of AI systems across various contexts. The general objective is to critically analyze the AI risk classification systems in the AIA and in Brazilian National Council of Justice Resolution No. 615/2025, proposing guidelines for more effective models. Specifically, the study aims to compare the adopted approaches, identify their limitations, and suggest methodological improvements.

The methodology employed involves bibliographic and documentary research, using a comparative analysis of regulatory frameworks from the European Union and the Brazilian National Council of Justice. This paper proposes adopting a risk assessment model that integrates the AIA with well-established approaches, such as the IPCC framework, which incorporates the dimensions of hazard, exposure, vulnerability, and response, analyzing the interactions among these components to estimate the magnitude of AI risk.

In parallel, in Brazil, Resolution CNJ No. 615/2025 advances the regulation of AI in the Judiciary by establishing a risk categorization and explicit prohibitions on the use of automated systems. Although this approach adopts the principle of governance proportional to risk, the initial responsibility for classification falls on the courts themselves, which may lead to biases and lack of uniformity.

The joint analysis reveals the need for dynamic and adaptive risk assessment models that go beyond static application-based categorizations. The incorporation of a more complex framework enables a more accurate analysis of AI risk magnitude in specific scenarios, considering multiple interactions and impacts on fundamental values. In the Brazilian judicial context, this implies strengthening external oversight mechanisms and ensuring the independence and technical capacity of the National Committee, to prevent institutional bias and promote transparency in risk classification.