AI Runtime Governance: A Multi-Layer Framework for Shadow AI Threat Detection in Enterprise Environments

The rapid proliferation of Large Language Models (LLMs) and autonomous AI agents in enterprise environments has created unprecedented security challenges collectively termed Shadow AI - the unauthorized or unmonitored use of AI systems that bypass corporate governance frameworks. This white paper introduces AI Runtime Governance, a novel security paradigm that addresses threats emerging from generative AI adoption, including prompt injection attacks, sensitive data exfiltration through AI interfaces, and the uncontrolled actions of autonomous agents operating beyond human oversight. We present the AI-Sentinel platform, Brazil's first integrated solution for AI Runtime Governance, which implements a multi-layer classification architecture combining heuristic pattern matching, regular expression analysis, and deep learning models (DeBERTa-based) to achieve real-time threat detection across browser, mobile, and network perimeters. The framework aligns with international compliance requirements including ISO 42001 (AI Management Systems), ISO 27001, LGPD, GDPR, and the EU AI Act. --- RESUMO (Português): A rápida proliferação de Large Language Models (LLMs) e agentes autônomos de IA em ambientes corporativos criou desafios de segurança sem precedentes, coletivamente denominados Shadow AI. Este white paper introduz o conceito de AI Runtime Governance, um novo paradigma de segurança para ameaças emergentes da adoção de IA generativa. Apresentamos a plataforma AI-Sentinel, a primeira solução integrada do Brasil para Governança de IA em Tempo de Execução, com arquitetura multicamada alinhada às normas ISO 42001, ISO 27001, LGPD, GDPR e EU AI Act. --- For collaboration, licensing, or technical demonstrations: Contact: paulocamara@epreserve.com.br | npctecnologia@gmail.com | www.linkedin.com/in/iasentinelcompliance | Santos, SP, Brazil