Environment-Grounded Multi-Agent Workflow for Autonomous Penetration Testing

ArXi:2603.24221v1 Announce Type: cross The increasing complexity and interconnectivity of digital infrastructures make scalable and reliable security assessment methods essential. Robotic systems represent a particularly important class of operational technology, as modern robots are highly networked cyber-physical systems deployed in domains such as industrial automation, logistics, and autonomous services. This paper explores the use of large language models for automated penetration testing in robotic environments.