From Deepfake Detection Research to Public Policy: A Survey on Generalization Challenges in Synthetic Media for the Brazilian Electoral Context

Deepfake detection has advanced rapidly in recent years, yet its practical effectiveness remains limited when models are exposed to real-world conditions that differ from controlled benchmarks. This gap is particularly critical in high-stakes scenarios such as electoral processes, where synthetic media can influence public perception at scale. This paper investigates the generalization limitations of current deepfake detection approaches, with emphasis on their applicability in the Brazilian context. Rather than focusing solely on algorithmic performance, we frame deepfake detection as a socio-technical problem, where issues of trust, interpretability, and user response are central to system effectiveness. Through a critical analysis of existing datasets, evaluation protocols, and detection strategies, we show that current research paradigms often fail to account for distribution shifts, cultural variability, and adversarial adaptation. By bridging detection research with policy-oriented perspectives, this work contributes to a more actionable understanding of how synthetic media challenges can be addressed in Brazil and similar socio-political contexts. Finally, we explicitly derive implications for public policy in the Brazilian electoral context, connecting observed technical limitations to challenges in regulation, platform governance, and the use of automated systems in institutional decision-making.