INTEGRATED APPROACH TO THREAT MODELING IN ARTIFICIAL INTELLIGENCE SYSTEMS

This paper substantiates the relevance of threat modeling for artificial intelligence (AI) systems in the context of increasing model autonomy and the emergence of new attack vectors. It demonstrates that traditional methods fail to account for the specific nature of AI, creating the need for a comprehensive approach capable of covering the entire system lifecycle. The methodological foundation of the integrated approach combines international standards and industry best practices: ISO/IEC 42001:2023 ensures governance and auditing, NIST AI RMF 1.0 defines the process cycle Govern–Map–Measure–Manage, MITRE ATLAS enriches models with realistic attack scenarios, CSA MAESTRO introduces multi-layer architectural decomposition, and OWASP GenAI Security Project provides operational artifacts and prioritization tools. This synthesis enables the integration of strategic policies, technical taxonomies, and practical playbooks into a single managed process. The proposed approach makes threat modeling continuous and evidence-based, ensuring traceability from threat identification to control implementation and performance metrics. It addresses both technical and socio-technical risks, includin