Abstract
This paper introduces a hybrid neuro-symbolic multi-agent architecture designed to strengthen the safety, interpretability, and reliability of autonomous systems. The proposed framework integrates neural perception and planning with symbolic reasoning, enabling agents to generate flexible action hypotheses while maintaining strict logical coherence through rule-based verification. The system is evaluated in robotic coordination and diagnostic environments, demonstrating improved task success, reduced unsafe behaviors, faster learning convergence, and clearer reasoning traces compared to neural-only and symbolic-only baselines. This approach offers a scalable foundation for next-generation autonomous systems requiring accountability, adaptability, and multi-agent consistency.