Violenza algoritmica di genere, tecnologie predittive e diritto internazionale

This article departs from a structural paradox in contemporary international law: the algorithmic risk assessment systems deployed by States to protect domestic violence victims can themselves become institutional mechanisms producing the very harm they were designed to prevent. From this premise, the article develops two original interpretive categories, «algorithmic gender violence» and «algorithmic nonrefoulement», and applies them to a systematic comparison of four European models (VioGén, SARA, DASH and the Italian Codice Rosso), assessed against a multilevel normative framework integrating CEDAW, the Istanbul Convention, the AI Act, the GDPR and the UN Guiding Principles on Business and Human Rights. Drawing on feminist-critical methodology, the analysis demonstrates that the structural tensions identified across each system reflect an unresolved contradiction between the probabilistic logic of predictive models and the demands of an international legal order requiring autonomous, non-discriminatory and victim-centred risk assessments. The article identifies six cumulative requirements of algorithmic compliance with existing international law and advances de lege ferenda proposals, including an Optional Protocol to CEDAW on digital gender-based violence and a binding algorithmic gender audit obligation with legal consequences for non-compliant States.