O poder de governar as companhias: a velha crise de confiança na era da Inteligência Artificial

Abstract

Traditionally, corporate governance has been concerned with guiding how companies are directed, in response to recurring crises of trust related to the definition of who should hold the power to control them and for what purposes such power is exercised. In the Brazilian legal framework, although Law No. 6,404/76 assigns to the general shareholders’ meeting the authority to deliberate on corporate matters and to the board of directors the general guidance of the company’s business, decision-making power proves to be structurally concentrated in the figure of the controlling shareholder. Even though legally regulated, this power finds both formal and informal means to influence decision-making bodies, directing them according to its own interests, sometimes to the detriment of the corporate interest and other stakeholders. Based on this premise, the present study investigates how this normative arrangement—marked by shortcomings in constraining control power—interacts with the incorporation of AI systems into corporate decision-making processes. The central object of the research is thus to understand this phenomenon in light of the power structures and corporate governance arrangements specific to the Brazilian context, given that the international literature on the subject has predominantly developed in settings characterized by dispersed ownership and conflicts between shareholders and managers. To this end, Part I presents a literature review grounded in the work of authors who critically examine the centrality of the controlling shareholder and the normative mechanisms aimed at limiting their powers, followed by an empirical analysis of administrative decisions issued by the Brazilian Securities and Exchange Commission (CVM) between 2014 and 2024. This analysis is conceived as a qualitative illustration of the adopted theoretical framework, with the purpose of identifying concrete correspondences regarding the influence of control power as a factor in destabilizing the functional autonomy of corporate bodies and in the practical reconfiguration of formally established decision-making arrangements. Building on this analytical foundation, Part II proposes an exploratory examination of the integration of AI systems into decision-making by corporate bodies, considering variables such as accuracy, transparency, risks, and levels of human oversight, according to the degree of technological autonomy—assisted, augmented, or autonomous. From this mapping, it is possible to infer that the greater the centrality of the decision and the more significant its effects on rights, duties, and the company’s strategic direction, the greater the need for proportional safeguards. In light of the Brazilian institutional and regulatory context, and in dialogue with concerns primarily formulated at the international level, the study concludes that the incorporation of AI systems into the deliberations of corporate bodies is not, in itself, a corrective element for the identified shortcomings. If implemented without consideration of previously identified distortions and the mismatch between the formal allocation of authority and the actual centers of control, technology tends to replicate—through a new operational language—the traditional forms of asymmetric exercise of power.