Towards agentic intelligence in digital twins: advances, use cases and challenges

Official URL: https://dx.doi.org/10.1109/MIOT.2026.3727905

Digital twins (DTs) are emerging as a key approach for managing the growing complexity of 6G and beyond wireless networks. However, they lack closed-loop autonomy, multi-step reasoning, and the ability to act independently in dynamic environments. At the same time, agentic AI, where autonomous agents can perceive, reason, act, and learn, is maturing rapidly. Accordingly, this paper examines the convergence of agentic AI and DTs. We first review the building blocks of agentic AI, including agent architectures, multi-agent coordination, communication protocols and autonomy levels, and map them to DT-based network management. We then propose an architecture that places agents inside containerized DT sandboxes for automated security assessment. To evaluate this approach, we benchmark six frontier large-language models across 18 scenarios covering seven vulnerability categories. Building on the results obtained, the study also discusses cost trade-offs, per-category performance gaps, and deployment recommendations for different operational needs. Finally, it identifies open challenges and future directions for scaling agentic intelligence in network DTs.