Detalhes do Documento

Financial institutions increasingly depend on machine learning and generative artificial intelligence to underwrite credit, detect fraud, price risk, execute trades, and serve customers through conversational agents. This growing reliance introduces an attack surface that traditional information-security controls and traditional model-risk-management (MRM) programs were never designed to address in combination. Adversarial perturbations, training-data poisoning, model extraction, membership inference, and prompt injection against large language models (LLMs) sit alongside conventional operational-risk concerns such as model drift, overfitting, and inadequate validation. This paper proposes an integrated framework, termed AI Security and Model Risk Governance (AI-SMRG), that fuses cybersecurity engineering with prudential model-risk practice into a single, continuously monitored discipline rather than two parallel compliance exercises.....