Enterprise AI Analysis

Unlocking Insights from Your Cyber Logs

Log data is crucial for intrusion detection and forensic investigations. Manual analysis is challenging due to high data volumes and heterogeneous formats. Automated methods exist but often rely on domain-specific configurations. Large Language Models (LLMs) offer domain- and format-agnostic interpretation but lack suitable public datasets for evaluation. This paper introduces CAM-LDS, a dataset of system log data and security alerts for LLM-based interpretation. It covers 81 distinct attack techniques across 13 tactics, collected in a reproducible Linux environment. The dataset includes log events, network captures, IDS alerts, and system configurations. A case study with an LLM shows correct prediction of attack techniques for approximately one-third of attack steps, demonstrating the potential of LLM-based log interpretation and the utility of the data set.