A group of OpenAI researchers ensuring that a future AGI system does not pose catastrophic or even existential danger has been decimated.