“Demis Hassabis has shifted his view from believing in a 'singleton scenario' for AGI development to now seeing AI safety as a collective action problem that only governments can solve.”