How are networks that have high gradients and operating in linear regime more vulnerable to adversarial attacks?

Question

At Stanford, the CS230 lecture mentioned that "high gradients and operating in linear regime more vulnerable to adversarial attacks?" Can someone explain how?

Video link: https://youtu.be/ANszao6YQuM?list=PLoROMvodv4rOABXSygHTsbvUz4G_YQhOb&t=2726