Four Horsemen of the AI-pocalypse

Machine learning systems bring powerful capabilities, but also introduce new, often overlooked cybersecurity risks. In this session organized by the Virtual Routes AI-Cybersecurity Research and Policy Hub, Dr. Sven Herpig explored how attacks can range from physical manipulations, such as altered lane markings, to technical exploits like neural network trojans and prompt injections. He explained why some of these threats may not yet be critical, and focus on what truly matters right now: the shift toward ML-first software design, the limits of traditional IT security measures, and the emergence of self-reinforcing AI “vibe coding” cycles that can amplify vulnerabilities.