Beyond Runtime Enforcement: Shield Synthesis as Defensibility Analysis for Adversarial Networks
Researchers propose a new approach to analyzing the defensibility of adversarial networks, shifting focus from runtime enforcement to design-time analysis. The method uses automata-theoretic machinery to construct a constrained two-player safety game, yielding a formal certificate of defensibility. This framework provides structural insights and topology-level metrics, capturing both formal safety properties and operational behavior under adaptive play, offering a more nuanced understanding of network security than traditional runtime constraints. AI
IMPACT This research offers a new framework for understanding and improving network security by analyzing defensibility at the design stage, potentially leading to more robust AI systems.