South Korea is updating its artificial intelligence security guidance as companies deploy AI agents capable of operating with limited human supervision, while Microsoft is proposing model-level rules requiring its systems to remain subject to human interruption and shutdown.