New AI risk paper maps extinction paths without malice

A new arXiv paper argues that advanced artificial intelligence could cause human extinction through systemic failure modes rather than conscious hostility or malicious intent. The analysis identifies four distinct pathways to catastrophe: autonomous misalignment, harmful human use, organizational failure, and competitive deployment pressures. Risk severity depends on specific technical factors including system autonomy, external access capabilities, and the preservation of societal recovery capacity. The authors emphasize that catastrophic outcomes do not require AI consciousness or explicit goals to harm humanity. This framework is deliberately non-operational, focusing instead on identifying causal conditions and empirically tractable defensive research questions. The paper aims to redirect safety discourse toward measurable intermediate quantities and institutional safeguards. By decoupling extinction risk from anthropomorphic assumptions, the analysis provides a structured basis for evaluating civilizational resilience against advanced AI systems.