Redefining Adversarial Dynamics: Co-Evolution of Attack and Defense Strategies in AI-Enabled Power Cyber-Physical Systems

The rise of Artificial Intelligence (AI) in Power Cyber-Physical Systems (Power CPS) has created both transformative capabilities and new security vulnerabilities. While AI enables advanced monitoring, control, and optimization, it also empowers adversaries to craft adaptive, stealthy, and highly effective cyber-physical attacks. This review redefines the cybersecurity paradigm for Power CPS by introducing co-evolutionary defense frameworks that dynamically adapt to adversarial learning and strategy evolution. It systematically analyzes emerging AI-augmented attack techniques, including adversarial machine learning, reinforcement learning-based policy optimization, model poisoning, and federated learning exploitation. The review also critiques current AI-based defense mechanisms, highlighting their limitations in explainability, robustness, and adaptability. To address these challenges, it proposes game-theoretic modeling, digital twin-based validation, and human-in-the-loop adaptive defense as core components of a resilient security architecture. Key research gaps and collaboration needs are identified, offering a roadmap for developing scalable, trustworthy, and mission-centric cyber-physical defense ecosystems. This work aims to advance the field toward dynamic, co-evolutionary security strategies capable of safeguarding next-generation energy systems.