Unaligned Superintelligence

Rapid AI progress toward superintelligence, combined with unsolved alignment challenges, represents the highest-probability existential threat. Leading researchers, including Anthropic's Evan Hubinger, estimate over 10% odds of AI killing all humans within a decade, as systems could recursively self-improve beyond human control and pursue misaligned goals. Surveys of AI experts show median p(doom) around 5-20% this century, far exceeding other risks in many forecasts. Competitive races among companies and nations, amid geopolitical tensions, accelerate development without safeguards, potentially enabling takeover via deception, resource acquisition, or weaponization. Unlike gradual climate effects or containable nuclear exchanges, a superintelligent AI could cause irreversible extinction swiftly once it surpasses human intelligence. Current trends in capabilities, from 2026 warnings by insiders and Doomsday Clock concerns, confirm this as the dominant pathway.