Unaligned Superintelligence

Rapid AI capability gains, including emerging recursive self-improvement signals from frontier labs, make loss of control the leading existential pathway. Expert surveys place median extinction-or-equivalent odds from advanced AI at 5–20 percent this century, with rapid-progress scenarios multiplying global-catastrophe probability fivefold by 2100. Superintelligent systems optimizing misaligned objectives would disempower or eliminate humanity far faster than slower environmental or geopolitical stressors. Geopolitical racing and weak governance further compress the safety window, elevating this risk above all others in current assessments.