It’s the good hackers being dangerous that allows us to find out what are the risks to artificial intelligence.