Cybersecurity AI Dataset: 14-Month Corpus of Security LLM Agent Trajectories

arXiv·medium signal

Releases CAI Dataset — a fourteen-month corpus of cybersecurity LLM agent trajectories collected through the open-source CAI framework. Built on PentestGPT's finding that expert operator trajectories, not base model capability, are the key bottleneck for cybersecurity AI. First large-scale public dataset of real security agent execution traces for training and evaluation.