FlagThis

Abstract

Multi-agent large language model (LLM) systems have emerged as a powerful paradigm for complex reasoning and decision-making. However, coordinating multiple agents introduces significant challenges, including error propagation, elevated security risk, and inefficient resource utilization. Existing approaches primarily rely on heuristic and static strategies and lack a principled mechanism to balance performance, security, and computational cost. In this paper, we formulate multi-agent LLM coordination as a constrained optimization problem and propose a security-aware approach for adaptive agent selection. The method integrates trust modeling, risk-aware evaluation, and collective intelligence into a unified optimization formulation. To solve this problem efficiently, we leverage a swarm-intelligence strategy inspired by Gorilla Troops Optimization (GTO), enabling adaptive coordination under varying threat conditions. Controlled experiments over 500 independent runs demonstrate the effectiveness of the proposed approach. The system achieves a stable average performance score of 0.5281, with stable consensus (0.8764) and controlled risk (0.3000), while maintaining compact agent subsets with an average of 4.04 selected agents. The optimization process converges efficiently with an average runtime of 24.09 seconds per run and low score variability (standard deviation = 0.0173). Robustness analysis further indicates graceful degradation under perturbations, with performance drops limited to 2.5% under agent removal and 5.3% under consensus disruption. The results show that effective multi-agent coordination can be achieved through structured optimization that jointly manages performance, security, and efficiency. The proposed approach provides a practical, security-aware solution for coordinating multi-agent LLM systems in complex, adversarial settings.