TwinPot-Gen: A Generative Semantic Digital-Twin Honeypot Framework for Adaptive and Explainable Defence in 6G Networks

The security of emerging 6G small-cell networks increasingly depends on intelligent deception mechanisms, where honeypots serve as proactive defence components. However, conventional honeypots lack semantic and behavioural realism and often risk leaking sensitive data when emulating complex network states. Existing systems remain limited by static personas, weak sanitisation, and unexplainable reactions, making them ineffective against adaptive attackers. To address this challenge, we propose TwinPot-Gen, a digital-twin-assisted honeypot framework that integrates semantic communication, graph-based knowledge modelling, and generative AI to achieve realistic and privacy-preserving deception. The framework establishes two isolated domains, the Digital-Twin Network and the Honeypot Network, connected through two secure gateways: the Generative Knowledge Gateway (GKG) and the Defence-Orchestration Gateway (DOG). GKG performs semantic sanitisation and persona generation using a large language model, while DOG enforces adaptive, policy-driven responses. Experimental results on a Kubernetes-based 6G testbed show that TwinPot-Gen improves persona realism by 35-50%, achieves a 0.94 F1-score in attack detection, and maintains sub-100 ms orchestration latency, confirming its efficiency and trustworthiness in dynamic 6G environments.