They've come up with plenty (e.g. https://blog.openai.com/adversarial-example-research/ ) - noone broadcasts them widely because noone likes the answers