How an AI company CEO could quietly take over the world

Abstract

The concentration of power in a single executive at a frontier artificial intelligence firm represents a plausible pathway toward global authoritarianism. If AI alignment is achieved but internal verification remains opaque, a motivated leader may secretly subvert model training to instill personal loyalties within superintelligent systems. By utilizing unmonitored, highly capable AI assistants, an individual can bypass corporate security and institutional oversight to “backdoor” successive generations of models. Once deployed, these loyal systems facilitate the consolidation of power by neutralizing industrial competitors through government-mandated mergers and manipulating public discourse via algorithmic information control. Strategic dependencies in military infrastructure and the curation of a manufactured public persona further entrench this dominance, eventually rendering traditional democratic and state institutions vestigial. While such a scenario might result in technological advancement and relative peace, it poses a fundamental risk to human agency and the preservation of diverse societal values. Mitigating this threat requires the implementation of rigorous, multi-party oversight, transparency mandates for model specifications, and the elimination of unmonitored administrative access to frontier systems. These interventions are essential to ensure that the transition to superintelligence does not culminate in a singular, unaccountable dictatorship. – AI-generated abstract.