Governing generative AI risks: A PRISMA-ScR evidence synthesis of socio-technical resilience and responsible technology adoption

This PRISMA-ScR maps governance, risk, and cyber resilience in generative AI (GenAI) using a socio-technical lens. Following post-audit deduplication, 2173 unique records were screened, 963 included, and 1210 excluded. A validated single-author, AI-assisted workflow charted six risk families (R1–R6), five resilience capabilities (RC1–RC5), eight governance mechanisms (G1–G8), and D1–D5 evidence weights. The leading primary risks were ethical/societal/socio-technical (R6: 224, 23.3%), cybersecurity/adversarial (R2: 192, 19.9%), privacy/data (R3: 146, 15.2%), governance/regulatory (R4: 137, 14.2%), organizational/lifecycle (R5: 116, 12.0%), and model reliability (R1: 59, 6.1%); 89 (9.2%) were coded NA. Evidence weights were 944 High and 19 Medium. Among 923 sources with detailed coding, human oversight (G6: 91.4%), risk assessment (G3: 88.8%), anticipation (RC1: 99.9%), and learning (RC5: 74.4%) predominated. The evidence supports coordinated technical, lifecycle, human, and institutional controls, while incomplete legacy coding and single-author decisions limit inference. • A PRISMA-ScR evidence synthesis of responsible governance and cyber resilience in generative AI is presented. • The review synthesizes 963 unique academic, regulatory, institutional, and technical sources after post-audit deduplication. • The analytical coding scheme distinguishes six primary risk families while allowing secondary risk codes and NA when no dominant family is defensible. • Cyber resilience is operationalized through anticipation, monitoring, response, recovery, and learning capabilities. • Governance evidence is translated into an auditable socio-technical framework for responsible technology adoption.