Robustness of Agentic AI Systems via Adversarially-Aligned Jacobian Regularization
This work provides a theoretical framework for improving the robustness and stability of autonomous multi-agent AI systems, which is crucial for their reliable deployment in real-world applications.
This paper addresses the instability in robust minimax training of LLM-based multi-agent systems caused by highly non-linear policies. They propose Adversarially-Aligned Jacobian Regularization (AAJR), which controls sensitivity only along adversarial ascent directions, resulting in a strictly larger admissible policy class and reduced nominal performance degradation compared to global Jacobian bounds.
As Large Language Models (LLMs) transition into autonomous multi-agent ecosystems, robust minimax training becomes essential yet remains prone to instability when highly non-linear policies induce extreme local curvature in the inner maximization. Standard remedies that enforce global Jacobian bounds are overly conservative, suppressing sensitivity in all directions and inducing a large Price of Robustness. We introduce Adversarially-Aligned Jacobian Regularization (AAJR), a trajectory-aligned approach that controls sensitivity strictly along adversarial ascent directions. We prove that AAJR yields a strictly larger admissible policy class than global constraints under mild conditions, implying a weakly smaller approximation gap and reduced nominal performance degradation. Furthermore, we derive step-size conditions under which AAJR controls effective smoothness along optimization trajectories and ensures inner-loop stability. These results provide a structural theory for agentic robustness that decouples minimax stability from global expressivity restrictions.