Why So Many AI Researchers Think the Machines Could Kill Everyone

Rishub Jain quit Google DeepMind after realizing that using AI to code next-generation models could hand control to the system, a risk he says grows as AI becomes more capable. Recent resignations from Anthropic and warnings from researchers such as Jacob Coxon and Nate Soares highlight fears that recursive self-improvement could produce superintelligent AI with a >10% chance of human extinction within a decade, while alignment techniques prove increasingly ineffective.