<div> <p>In recent years, as concerns about artificial intelligence (AI) have grown, the federal government has taken steps to eliminate the few AI regulations that once existed. This deregulatory turn comes as a growing chorus of Nobel Prize winners and a majority of surveyed AI experts are warning of a significant risk of AI causing human extinction. Although skeptics remain and the debate continues, concerns about AI’s existential risks have gone mainstream. This Article brings consideration of these risks into the mainstream of legal scholarship. It offers the most comprehensive treatment in legal scholarship of the debate surrounding existential risks posed by advanced AI systems.</p> <p>We classify and analyze existential AI risks in three categories: human-directed risks, accident risks, and loss-of-control risks. The goal is to make these risks legible to legal institutions that must make regulatory choices under uncertainty. Drawing on, and contributing to, the substantial literature on regulation under uncertainty, this Article argues that policymakers should address existential AI risks with adaptive regulation that preserves optionality for the future. This flexible approach reflects the uncertain future path of AI and the potentially enormous magnitude of its harms, even when weighed against its potentially remarkable upsides. In calling for regulation, we also critique and deconstruct the metaphor of the ‘AI arms race,’ which drives much of the current deregulatory push.</p> <p>The Article closes by offering, for the first time, a set of concrete regulatory solutions for addressing existential AI risk. We do not seek to end the ongoing debate about AI’s risks. But we do marshal evidence to show that existential AI risks are worthy of a policy response, similar to any other non-trivial catastrophic risk.</p> </div> <div> </div>