Robust Adversarial Examples

OpenAI created images that reliably fool neural network classifiers when viewed from varied scales and perspectives challenging a claim from last week that self-driving cars would be hard to trick maliciously since they capture images from multiple scales, angles, perspectives, and the like.