Who is liable when no one has decided? Civil liability for algorithmic harm between the Code, the regulatory framework and the Supreme Court

Abstract

This article examines whether the reform of the Brazilian Civil Code resolves or exacerbates the competition among liability regimes applicable to harm caused by artificial intelligence systems. To that end, it employs a qualitative research design combining comparative documentary analysis and content analysis, applied to four sets of documents, with the corpus limited to materials available through 31 December 2025: Bill No. 4/2025; the engrossed text of Bill No. 2,338/2023; the judgments handed down under General Repercussion Themes Nos. 987 and 533; and the records of the eight public hearings held by the Federal Senate’s Temporary Committee. The simultaneous examination of the three regimes, grounded in the records of the legislative and judicial proceedings rather than in scholarly commentary on the bills, makes it possible to observe the emergence of competition among normative frameworks, rather than merely its outcome. The findings indicate that the Civil Code reform bill contains a mandatory coordination clause linked to the regulatory classification of risk, but that the clause is misdirected. The engrossed text combines statutory high-risk categories with regulatory functions assigned to the competent authority and the sectoral authorities, while Chapter V preserves the applicability of the Brazilian Consumer Protection Code, the Civil Code, and special legislation without establishing an autonomous regime of risk-graded strict liability. What emerges, therefore, is a mismatch between evolving normative architectures, rather than a conflict between rules currently in force. The article proposes a comparative liability matrix structured around six categories, as well as a distinction between a coordinating function and an exhaustive regulatory function, from which three drafting implications for Book VI follow.