Policed by Code: AI, Gender and Justice in the Global South.

Abstract

Algorithmic policing systems in the Global South do not simply reproduce existing inequalities, they intensify and reorganise them. Evidence from three urban contexts, Hyderabad (India), Lagos (Nigeria), and Rio de Janeiro (Brazil), shows that AI-driven surveillance and predictive policing tools disproportionately target women situated at the intersections of gender, race, caste, and class. The central problem is not technical failure. It is the structural embedding of structural inequality within systems presented as neutral, efficient, and objective.

Grounded in intersectional feminist theory (Crenshaw, 1989) and postcolonial critiques of technology (Benjamin, 2019; Mbembe, 2003; Quijano, 2000), this paper develops the Algorithmic Harm Cascade, a five-stage model that traces how historical data inequalities evolve into contemporary forms of exclusion. The cascade follows a progression from discriminatory data inheritance to proxy encoding, deployment amplification, and governance absence, culminating in structural exclusion reinforced through behavioural adaptation. This framework distinguishes between performance bias, uneven technical outcomes, and sociotechnical bias, where systems function as designed but reproduce injustice. The latter emerges as the dominant form of harm across all three cases.

Analysis of over 250 publicly available sources, including peer- reviewed research, procurement records, civil society reports, and investigative journalism, reveals three consistent pathways through which inequality enters algorithmic systems: the incorporation of biased enforcement records into training data, the encoding of class and gender through proxy variables, and the construction of bodily visibility as a basis for suspicion.

Building on these findings, the paper advances a three-part governance framework, Power-Aware AI Systems, Memory Justice, and the Right to Algorithmic Refusal, grounded in empirical realities rather than abstract ethical commitments. Across all cases, governance capacity remains limited, with no city exceeding a score of 2.5 out of 5 on key regulatory dimensions, underscoring a widening gap between technological deployment and institutional accountability.

Keywords: artificial intelligence policing, gender bias, intersectionality, surveillance, postcolonial technology, algorithmic justice, sociotechnical bias, governance, Global South