In the context of smart cities, the use of intelligent surveillance systems has grown significantly, raising both opportunities for enhanced public safety and challenges related to ethical digital practices. Anomalous activity recognition (AAR) is critical for preventing violent incidents and ensuring rapid emergency response. However, current approaches often rely on isolated spatial or temporal features and overlook broader ethical responsibilities such as transparency, fairness, and responsible deployment in public environments. This study presents a socially responsible, deep learning–based surveillance framework that aligns with the principles of digital social responsibility (DSR). The proposed system integrates a multi‐tier residual network (MTRN) for extracting salient spatial features from complex visual scenes and a parallel residual temporal network (PRTN) for modeling behavioral dynamics across time. Together, they enable robust, context‐aware detection of abnormal events in surveillance footage. Crucially, the framework incorporates a real‐time alert mechanism designed not only to assist law enforcement but also to promote ethical response protocols that minimize harm and safeguard civil liberties. Empirical evaluations on the RWF‐2000, Surveillance Fight, and Industrial Surveillance datasets demonstrate high accuracy (97.40%, 96.35%, and 81.70%, respectively), highlighting the model’s reliability and real‐world applicability. By contributing to safer urban environments while upholding principles of fairness, inclusivity, and accountability, this work exemplifies the role of ethical AI systems in building trustworthy and sustainable digital societies. The code will be available at the Samee‐ARIC GitHub repository.
Khan et al. (Thu,) studied this question.