Legally-aware AI moderation in Vietnam: A cybersecurity law-compliant framework for platform governance

Digital platforms increasingly mediate fraud, illicit trade, misinformation, and incitement, often under stringent statutory timelines that demand rapid and auditable responses. Conventional moderation pipelines remain tuned to classifier accuracy, leaving them misaligned with legal categories, due-process safeguards, and deterrence goals. This produces costly false positives that chill lawful expression, false negatives that enable social harm, and thin audit trails that weaken accountability. We address this gap by introducing a legally-aware AI moderation framework that integrates law → policy → model label → evidence minima → action tier → appeal path into a single, auditable pipeline with human-in-the-loop review and mandatory rationales. The framework embeds economic criminology by modeling the expected utility of offending and demonstrating how risk thresholds, graduated sanctions, and appeal quality shift incentives while internalizing enforcement externalities. We develop a decision-analytic simulation of 50,000 multi-signal items that calibrates governance choices on a harm–cost frontier. Results show that at matched thresholds, a graduated action menu achieves 2.51× greater harm reduction per unit enforcement cost compared to binary removal, while sustaining deterrence and lowering appeal overturn rates. This evidence demonstrates that moderation choices function as price signals in a crime market: raising certainty and proportional severity reduces offenders’ expected payoffs while protecting legitimacy through reversibility and transparency. By quantifying deterrence, externality reduction, and enforcement-cost efficiency, the framework reframes moderation as a problem of governance economics, not just technical accuracy. Situated within economic criminology, it operationalizes deterrence (certainty × severity), rational-choice incentives, and capable guardianship via a cost-adjusted harm-reduction frontier. Through open research objects and a jurisdictional portability protocol, it offers a replicable and auditable blueprint for platform governance that is compliant, economically rational, and transferable across sectors and jurisdictions.