A novel risk assessment model for assessing generative AI risks in sustainable manufacturing: Insights from SME machinery firms

Generative artificial intelligence (AI) is increasingly adopted in manufacturing to improve operational efficiency and production economic performance, but its deployment also introduces risks, particularly for small and medium sized machinery firms (SMEs) with limited resources, legacy assets, and reliance on external services. A key managerial issue is how to balance efficiency oriented AI adoption with effective risk control in sustainable manufacturing. This study develops an uncertainty aware risk assessment framework integrating an interval cloud model, improved ITARA weighting, and improved probabilistic grey relational analysis. The framework extends Failure Mode and Effects Analysis (FMEA) by augmenting severity, occurrence, and detectability with cost of protection, impact on sustainability goals, and risk propagation. Linguistic expert ratings are mapped to expectation, entropy, and hyper entropy, while subgroup interval bounds are retained to avoid premature information loss. The approach is demonstrated in Taiwanese SME machinery enterprises using a panel of 24 senior practitioners, yielding 32 failure modes across technical, organizational and human, operational, supply chain, and legal, regulatory, and ethical dimensions. Detectability receives the highest weight, indicating the importance of early identification and containment. The highest risk modes are cybersecurity vulnerabilities from third party AI tools, institutionalized incorrect knowledge, and observability gaps on legacy equipment. Sensitivity analyses, including an epsilon sweep and leave five experts out replications, show stable criterion weights and an invariant top ten set. The results support a staged governance pathway that first reduces external exposure, then verifies knowledge and data, and finally strengthens shop floor observability and capability alignment for sustainability-oriented adoption.