Search for a command to run...
Abstract Nick Bostrom’s influential typology of information hazards defines such hazards as risks arising from the dissemination of true information. This truth-only restriction has shaped subsequent work on biosecurity, artificial intelligence, and existential risk, where the focus has fallen on technically actionable, capability-enhancing discoveries. This paper argues that the restriction is both conceptually too narrow and subtly self-undermining. Once “truth” is tacitly approximated by practical applicability, large classes of soft-theoretical and fictional representations—including those that structure debates about consciousness, moral status, and future risk—are relegated to the margins, even though they guide action and evaluation. I propose an alternative starting point: a notion of representation hazards, in which the primary units of analysis are representational regimes—beliefs, models, narratives, myths, and postulatory entities—whose uptake or disruption predictably alters agents’ vulnerability to harm. Within this framework, Bostrom-style information hazards appear as a special case involving true, capability-enhancing representations, while constructs such as qualia and fictional or therapeutic narratives are recognized as central instances of hazard-relevant representations. Using qualia as an example of postulatory entities and fiction as a test case for both hazardous and protective representations, the paper sketches how a representation-centered ontology can better capture the mixed ecology of fact and fiction in which hazard assessment actually occurs. It concludes by discussing implications for governance, disclosure norms, and the reflexive status of information-hazard typologies themselves. Information Hazards, Contradiction, Qualia, Representation Hazards, Nick Bostrom AI collaborator statementThis work was developed in sustained collaboration with OpenAI’s ChatGPT (GPT-5.1 Thinking), which served as a structural refiner (proposing and revising the overall organization of the framework), a stress-tester (probing for inconsistencies, gaps, and failure modes), and a stylistic drafter (producing initial text later edited and curated by the human author). The human author remains the sole legal and moral author of record, responsible for the selection, acceptance, and modification of all AI-generated material and for all substantive claims made in this document. New version purely for PDF format.