As artificial intelligence (AI) nears human-level performance in more complex cognitive tasks (Bubeck et al., 2023; Chollet, 2025), it becomes increasingly urgent to consider what may happen once AI surpasses the most intelligent humans. Artificial superintelligence (ASI) may seem distant, but accelerating progress in AI (Krizhevsky et al., 2012; Vaswani et al., 2017; Kaplan et al., 2020; OpenAI, 2022-2026) suggests it could arrive sooner than many anticipate. Given the asymmetrical risks posed by underestimating the rate of future progress, we start by outlining why one might cautiously expect ASI to be developed between 2029 and 2033. We then contend that existential risks may increase significantly during the development and deployment of ASI, such that through the 2030s, existential risk may reach its highest levels since the 1960s. In response, we introduce Superintelligence Challenges and Existential Risks as a framework for ASI risk management, where we focus on global challenges entailed by ASI, and map out failure modes traceable to each. We examine five main ASI challenges (technical safety, concentration of power, international governance, social health, and ethical challenges) where critical failure of any could result in catastrophic or existential risks. Our goal is to begin grounding ASI existential risks in a minimal set of overarching challenges that seem unavoidable.
Matthew Baxter (Sun,) studied this question.