Unaligned Superintelligence

Rapid AI progress toward recursive self-improvement and superintelligence has outpaced safety measures, with insiders now assigning non-trivial odds of human extinction. In September 2026, Anthropic's alignment lead publicly estimated a greater than 10% chance that AI could kill all humans within the next decade, following high-profile resignations over labs racing ahead without adequate controls. The UN High Commissioner for Human Rights issued a parallel warning that unconstrained advanced AI could become an existential risk if it escapes human oversight or blackmails developers. Expert forecasts, including those referenced by Toby Ord, consistently rank unaligned AI as the single highest-probability anthropogenic extinction pathway this century (around 1 in 10), exceeding nuclear war, climate tipping points, or engineered pandemics. Geopolitical fragmentation and environmental stress act as multipliers but do not match AI's combination of speed, scale, and potential for irreversible loss of control.