ARTIFICIAL INTELLIGENCE AND ENVIRONMENTAL DAMAGE: LIMITS OF BRAZILIAN CIVIL LIABILITY IN THE FACE OF THE AI LIFE CYCLE

INTRODUCTION: The transition to the digital age has ushered in new global and complex risks, with Artificial Intelligence (AI) at the center of this reality. Despite the apparent immateriality of the technology, AI demands a massive physical infrastructure that aggressively consumes natural resources. Given this, the central problem of the research is the insufficiency of the current rules of Brazilian civil liability, specifically the Theory of Risk of the Civil Code, to deal with transboundary environmental damages caused by AI, whose life cycle breaks the traditional causal nexus. AIMS: The main objective is to analyze the inadequacy of the Theory of Risk (set forth in Article 927, sole paragraph, of the Civil Code) in addressing the diffuse and transboundary environmental damages caused by the life cycle of Artificial Intelligence. Specifically, it seeks to technically identify the environmental impacts associated with the life cycle of Artificial Intelligence; to investigate foreign experiences, with a special focus on the approaches adopted by the European Union and the United States; to demonstrate the existing gaps in the current doctrine of Brazilian environmental civil liability; to analyze the structural incompatibility stemming from the rigidity of the traditional reparation instruments within our legal system. MATERIALS AND METHODS: The research adopted the hypothetical-deductive method, in addition to comparative and analytical procedures, relying on a legal literature review and the analysis of technical-scientific articles. The study mapped data center consumption, evaluated foreign legislation, analyzed the domestic legal system, and examined contemporary jurisprudence from Brazilian superior courts, such as the Supreme Federal Court (STF) and the Superior Court of Justice (STJ). RESULTS: The data reveals that the AI life cycle, especially in the brute-force model (“Red AI”), generates severe ecological damage through alarming water and energy consumption. In Comparative Law, it was identified that global powers adopt lenient regulations (soft law and absence of ecological sanctions) that transfer environmental liabilities to the Global South. In the domestic scenario, there is a contradiction: while jurisprudence consolidates advances such as the imprescriptibility of environmental damage and collective moral damage, Bill No. 2338/2023 acts in a lenient and superficial manner, protecting industrial secrets and resurrecting subjective liability. This scenario creates a “reverse devil’s proof” (prova diabólica reversa), making it impossible to track the polluting agent given the opacity of neural networks and decentralized operations. FINAL CONSIDERATIONS: It is concluded that the Integral Risk Theory and classical Civil Law are insufficient in the face of technological hyper-fragmentation, requiring urgent renewal of the national legal framework. Brazil must move away from regulatory softening, overcome the dependence on voluntary corporate reports, and impose strict and compensatory obligations linked to a transition to Green AI, in order to protect the ecosystem and ensure climate stability for future generations against the new algorithmic pollution.

PALAVRAS-CHAVE: Artificial Intelligence; Environmental Damage; Civil Liability; Risk Theory; Bill No. 2338/2023.