What solutions for managing risks on AI?

What are the main risks associated with the use of AI? And how can these risks be managed, so that the full potential of artificial intelligence systems can still be exploited?