RISK-BASED LIFECYCLE MANAGEMENT MODELS FOR TRUSTED ARTIFICIAL INTELLIGENCE SYSTEMS IN CRITICAL SECTORS

Tyryshkin Sergey Yuryevich, Ignatova Elena Ivanovna, Parfenov Pavel Dmitrievich

In critical sectors, artificial intelligence increasingly becomes part of operational and safety control loops, yet its behavior is shaped not only by software code but also by data, training procedures and the runtime environment. This makes trust assurance non-trivial and calls for a shift from fragmented checks to risk-based lifecycle management. The paper proposes a lifecycle model for trusted AI systems that integrates ISO/IEC 42001-aligned governance (GOST R ISO/IEC 42001-2024), AI system life cycle processes (GOST R 71539-2024), AI risk management guidance (ISO/IEC 23894:2023), the NIST AI RMF 1.0, and AI security practices (MITRE ATLAS, OWASP LLM Top 10). The suggested «evidence-based trust loop» includes criticality classification, stage-specific risk registers, data quality controls, independent validation, drift monitoring and change management. Practical examples from healthcare, energy and financial compliance illustrate how the model reduces hidden failures, model tampering and compliance risks through standardized assurance gates and auditable artifacts.

Keywords:trusted AI; risk management; life cycle; critical infrastructure; audit; verification and validation; data quality; model drift; MLOps; cybersecurity