Jueces algorítmicos y responsabilidad del operador judicial: límites jurídicos a la inteligencia artificial en la función pública y judicial. Estudio comparado entre Colombia y Brasil

Abstract

The incorporation of artificial intelligence systems into public administration and the judiciary has produced a remarkable normative convergence in Colombia and Brazil: both legal systems declare that a core of decision-making cannot be delegated to an algorithmic system. This article argues that such a declaration, on its own, amounts to a nominal guarantee. Through a documentary and case-law analysis framed within the study of transnational administrative convergence, along the lines of the global administrative law of Kingsbury et al. (2005), it examines judgments T-323 of 2024 and T-067 of 2025 of the Colombian Constitutional Court, Resolution 615 of 2025 of the Brazilian National Council of Justice, and article 20 of Brazilian Law 13.709 of 2018, set against three European decisions that have already adjudicated public algorithmic systems. The findings show that the human reserve operates in both countries at the final moment of the decision rather than across the system's life cycle, and that the two doctrinal routes employed, namely access to public information in Colombia and data protection in Brazil, yield unequal levels of enforceability, to the point that the Brazilian legislature removed in 2019 the requirement that review of automated decisions be carried out by a natural person. The analysis of two cases on the liability of judicial officers reveals that attribution is triggered by the officer's conduct and not by the system's behaviour. It concludes that the human reserve becomes enforceable only when accompanied by documented life-cycle traceability and a determinate regime of attribution.