Emerging technologies are often accompanied by catastrophic predictions about their likely effects (Winner, 1984). In the 1970s, scientists worried that recombinant DNA research might expose humanity to carcinogenic bacteria; in the 1980s, scientists feared that self-replicating nano-factories could turn the earth into formless ‘gray goo’; in the early 2000s, advances in Synthetic biology raised concerns about terrorists ordering smallpox in the mail; and in the 2010s, fears were raised that artificial intelligence might turn us all into paperclips. So far, however, none of these harms has come to fruition. Despite this poor track record of predictions, catastrophic narratives continue to play an important role in the governance of emerging technology. A role that imposes substantial costs on society. The main contribution of this paper is to extract from this history epistemic practices that explain the recurrence of catastrophic narratives and the limitations of existing proposals for addressing them.