Will AGI Systems Undermine Human Control? OpenAI, UC Berkeley & Oxford U Explore the Alignment Problem