uri-eliabayev talks with ben-mashiach, Director of Data & AI at Claroty, which secures cyber-physical systems: medical devices in hospitals, factory robots, and data-center power and cooling. The data science core is asset identification: building a “digital twin” inventory from passive traffic inspection and active queries across hundreds of protocols, then classifying a long tail of rare devices and extracting entities (model, serial number, OS version). That lets a hospital, for example, automatically segment what a newly connected device may reach. Classification is surprisingly hard: hierarchies don’t split cleanly, and an Xbox and a surgical robot can look alike until the last clue. So subject-matter experts and data scientists co-build features (expert fingerprinting techniques fed to models), with a tool that highlights unclassified regions of latent space for experts to label. LLMs help by embedding assets semantically for policies and search. Claroty also launched “Clar”, an agentic assistant built on the company’s accumulated expertise that can take security actions, not just report. AI lowers the barrier to attack, which makes knowing your own network’s baseline more important than ever.
Show: hidden-layers
Concepts: ai-cybersecurity · decision-classifiers · second-brain