Powerful AI systems approaching human-level intelligence (AGI) may arrive within years, but current secrecy in corporate and government labs risks catastrophic misalignment or authoritarian control.