Right from the Start: Dealing with AI Risks