Experts push back on doomsday AI claims, citing scant evidence

Recent high-profile warnings about AI posing existential threats are based more on sensational narratives than on concrete evidence, according to researchers. Mhairi Aitken of Our AI Collective argues that incidents like AI hacking tests reflect inadequate safeguards by companies, not rogue models. Andrew Rogoyski of the University of Surrey notes that the actual risk level has not shifted dramatically in recent years, despite periodic hype.

The article highlights that repeated AI alarmism often stems from high-profile figures rather than verifiable incidents. Researchers point to specific examples, such as AI hacking tests, where failures reflect missing corporate safeguards, not autonomous model behavior. Claims of recursive self-improvement or imminent superintelligence lack empirical support, and risk assessments have remained largely stable despite recent technical advances in mathematics and cybersecurity.

The piece also suggests that tech leaders’ calls for slowdowns may serve strategic interests, deflecting regulation while sustaining investor excitement. Former employees’ dramatic predictions are rarely countered by their employers, implying these narratives benefit corporate hype cycles. Meanwhile, the human tendency to project fears onto AI systems—seeing them as mirrors—amplifies public anxiety without corresponding evidence of actual danger.