Superhuman AI Bots Are Surprisingly Vulnerable to Exploits

This was originally published on post
By learning exploits from adversarial AI, people could defeat a superhuman Go-playing system