Threat Modelling for Artificial Intelligence Governance, Integrating Ethical Considerations into Adversarial Attack Simulations for Critical Infrastructure using Generative AI

As artificial intelligence (AI) becomes increasingly embedded in critical infrastructure, the risks of adversarial attacks on AI-driven systems have heightened concerns over security, governance, and ethics. Traditional threat modeling frameworks, while effective for conventional IT systems, are insufficient to capture the dynamic and evolving risks introduced by AI, particularly generative models capable of simulating sophisticated attack vectors. Addressing these gaps requires a governance framework that integrates both technical and ethical dimensions into adversarial risk assessment. This study explores a novel approach to threat modeling that embeds ethical considerations directly into the simulation of adversarial attacks against AI systems supporting critical infrastructure. It proposes a governance-oriented model in which generative AI is leveraged to replicate potential attack scenarios such as data poisoning, model inversion, and evasion while incorporating normative frameworks that assess impacts on fairness, accountability, and societal trust. By situating ethics alongside technical defenses, the approach ensures that mitigation strategies not only strengthen system resilience but also align with principles of responsible AI deployment. Case illustrations from energy grids, financial systems, and healthcare infrastructure demonstrate how generative AI-driven adversarial simulations can inform proactive governance, improve compliance with regulatory standards, and foster transparent risk communication. The results suggest that integrating ethics into threat modeling produces dual benefits: advancing resilience against malicious actors and embedding legitimacy and trustworthiness into AI governance for critical sectors.