OpenAI warns of ‘potentially catastrophic’ risks from superintelligent AI, outlines global safety measures