Exploring the Use of LLMs to Understand Network Traces

The growing threat of cyberattacks, combined with the increasingly complex digital landscape, has created a pressing need for cutting-edge cybersecurity solutions. The recent surge in large language models (LLMs) has brought both opportunities and risks, as their advanced language processing abilities can be harnessed for both good and evil purposes. In most cases for an organization, the network is the main gateway where security issues appears. The main problem with this issue is that the traffic of a big network is impossible to analyze by a human, that is why some of the tools used in networks analysis uses deep learning techniques to predict and solve problems. With this in mind and the recent success of LLMs for several types of tasks is the perfect motivation to check how good are the main models solving this hard job. For this purpose, in this paper we design a methodology to evaluate the ability of LLMs models to understand network traces using different prompt engineering techniques.