An Open, Bug Bounty-Ready Cybersecurity AI.pdf

This paper presents CAI, an open-source, agent-based Cybersecurity AI framework that automates planning, scanning, exploitation, and mitigation for offensive and defensive tasks. The authors demonstrate CAI by performing network reconnaissance, exploiting default credentials and exposed services (including a MIR-100 robot and web/FTP targets), running CTF and Hack The Box benchmarks where CAI outperforms or matches human teams in many tasks, and validating real-world bug-bounty findings; the work highlights CAI's capabilities, limitations, HITL role, and implications for vendor security claims and robot cybersecurity.