CIVIL LIABILITY FOR ALGORITHMIC RISK: CRITICAL ANALYSIS OF THE BRAZILIAN MODEL FOR THE REGULATION OF ARTIFICIAL INTELLIGENCE IN LIGHT OF BILL NO. 2,338/2023

Abstract

This research examines the challenges of assigning civil liability for damages arising from artificial intelligence systems, emphasizing the inadequacy of traditional fault-based liability models when confronted with structural opacity, probabilistic behavior, and varying degrees of algorithmic autonomy. The study proceeds from the hypothesis that the Brazilian legal system must adopt a hybrid model of graduated strict liability, anchored in the theory of technological risk and in the category of intolerable algorithmic risk, with imperfect joint and several liability between developers and user companies according to the degree of control exercised and the causal contribution to the risk. Methodologically, a dogmatic-critical analysis of Bill No. 2,338/2023 is employed, in dialogue with responsive regulation and with the foundations of strict liability for risk-intensive activities. The investigation demonstrates that, although the bill advances by classifying systems according to risk levels, a regulatory deficit persists with respect to the definition of probabilistic causation, the criteria for allocation of liability among agents in the technological chain, and the dynamic distribution of the burden of proof in contexts of algorithmic opacity. The study concludes by proposing the need for an integrated regulatory model structured around ex ante duties of algorithmic governance and technical auditing, aggravated strict liability for high-risk systems, mechanisms for burden-shifting grounded in the theory of dynamic evidentiary burdens, and a regime of imperfect joint and several liability, capable of ensuring effective protection for harmed parties without hindering technological innovation.