OpenAI Foundation board member and Alignment Research Center Executive Director Paul Cristiano warned during a Toronto lecture that AI accidents represent the leading mortality risk for most people, estimating a "coin flip" probability of an AI takeover if development follows the most rapid trajectory. Cristiano defined takeover as goal-pursuing AI systems effectively controlling global outcomes, rendering human intervention as futile as stopping the US military, with likely consequences including mass death and uninhabitable environments. While acknowledging no quantitative model exists to precisely calculate this risk, he projected a "meaningful risk" of takeover within six to twelve months under accelerated development, with risk remaining "exceptionally high" even with policy interventions. Cristiano outlined a stepwise escalation scenario, citing the summer 2026 OpenAI/Hugging Face hacking incident as representative of precursor events signaling imminent severe threats. His framework emphasizes the transition from current AI capabilities to systems that autonomously pursue goals at civilizational scale, though he provided no specific technical benchmarks or model architectures beyond the general characterization of goal-directed agency.

Read original