These types of ML approaches seem to always break down in the face of adversarial input. Security folks are going to have a field day if we’re not careful to m

These types of ML approaches seem to always break down in the face of adversarial input. Security folks are going to have a field day if we’re not careful to make sure that people really understand the limitations of these types of systems.