The State That Cannot See Its Own Tools: Algorithmic Opacity and the Limits of Transparency Law in Brazilian Public Administration

This article examines the transparency and accountability challenges raised by the use of AI systems in Brazilian public administration, using the Smart Sampa facial recognition program in the city of São Paulo as a central case study. Drawing on the work of Lessig (1999) and Crootof & Ard (2020) and integrating the multi-level transparency framework developed in public administration scholarship (Giest & Grimmelikhuijsen, 2020; Meijer & Grimmelikhuijsen, 2020), the article diagnoses Smart Sampa's opacity at three distinct levels: algorithmic, organizational, and institutional. It then examines six regulatory frameworks through the same analytical lens, identifying structural gaps common to the EU AI Act, the French Digital Republic Act, the Indian AI Governance Guidelines, and the three most relevant Brazilian legislative proposals. Integrating empirical evidence from citizen preference research (Feng & Chandra, 2025) and experimental studies on institutional trust (Grimmelikhuijsen & Meijer, 2022; Wenzelburger et al., 2024), the article argues that passive transparency mechanisms, whether algorithm registers, explanation rights, or formal access to information channels, cannot in themselves generate the accountability needed to govern high-risk AI systems in Brazilian public security contexts. The article proposes an institutional framework requiring pre-deployment review by a specialized algorithmic gatekeeper within the Judiciary, mandatory multi-level disclosure obligations, administrative appeals for affected individuals, and formal override of trade secrecy claims against gatekeeper access requests. The proposals do not require new primary legislation but the amplification of roles and powers already available in the Brazilian legal system.