A Google DeepMind research paper warns that Artificial General Intelligence (AGI) could emerge as early as 2030 and may carry risks capable of "permanently destroying humanity", though it stops short of detailing how such an outcome could unfold.

Co-authored by DeepMind’s Shane Legg, the paper classifies advanced AI risks into misuse, misalignment, mistakes, and structural issues, and emphasizes a proactive approach to prevent AI from being used to cause harm.

DeepMind CEO Demis Hassabis has also advocated for a global oversight body—similar to the UN or CERN—to ensure AGI development remains safe, transparent, and internationally regulated.