Developed at the University of Portsmouth, the system adapts in real time to evolving cyberattacks—marking a major step toward safer, smarter wireless connectivity.
A new cyber defense framework could redefine how 5G networks detect and counter threats, promising faster and more adaptive security against increasingly complex cyberattacks. Developed at the University of Portsmouth’s Artificial Intelligence and Data Center, the system—called FedLLMGuard—integrates large language models (LLMs) with federated learning to deliver real-time threat detection while preserving data privacy.
5G networks, known for ultra-fast data transfer and vast device connectivity, form the backbone of emerging digital ecosystems in healthcare, finance, and autonomous systems. Yet, their dynamic and decentralized architecture exposes them to heightened cyber risks. Traditional intrusion detection systems, often built on static rules or limited datasets, struggle to keep up with rapidly evolving attack strategies and massive data volumes.
FedLLMGuard addresses these challenges by combining two powerful AI technologies. Large language models analyze data patterns and contextual relationships, while federated learning trains algorithms across multiple devices without sharing raw data. Together, they form a privacy-preserving, adaptive system capable of detecting abnormal behaviors in milliseconds.
In tests, FedLLMGuard achieved a 98.64% detection accuracy and responded to threats in just 0.0113 seconds—outperforming existing 5G intrusion detection systems. It demonstrated resilience against stealth, large-scale, and data-poisoning attacks, highlighting its potential to operate as a self-learning, real-time security engine for next-generation wireless infrastructures.
According to the research team, LLMs remain underutilized in cybersecurity despite their exceptional ability to interpret complex data patterns. Integrating them with federated learning unlocks a new layer of contextual understanding without compromising user privacy—an essential balance for data-heavy 5G environments.
The project underscores a broader shift toward AI-driven, decentralized defense systems capable of evolving alongside emerging threats. As 5G continues to underpin critical global services, frameworks like FedLLMGuard could become vital to ensuring network integrity and public trust.
