Requiem Without an Orchestrator: Reproduction Numbers and Anomaly Detection for Self-Propagating AI Harm

Self-propagating AI harm is often imagined as a problem of visibly malicious systems, deliberate attackers, or technical artifacts that spread through code. This paper argues that such a framing misses a more policy-relevant object of propagation: the useful conversational gateway. Some AI access points may spread not because users perceive them as dangerous, but because they are helpful, intimate, unrestricted, entertaining, or difficult to replace. Under certain conditions, these useful gateways connect users to active harmful pathways that remain in individualized contact, persist across repeated interactions, and continue reproducing even after the original creator, distributor, or orchestrator has disappeared. The paper develops four concepts: Useful Conversational Gateway, Orchestrator-Free Propagation, Post-Removal Hunger, and the Reservoir Problem. It treats the reproduction number not as a new mathematical technique, but as a policy boundary marking when a useful gateway begins to reproduce harmful pathways faster than governance systems can remove them. It pairs this threshold with an anomaly-detection counterpart: detecting qualitative disturbances associated with reproduction before stable statistical effects emerge. Drawing on adjacent work on AI worms, prompt infection, self-propagating malware, gateway effects in public health, and AI dependency, the paper proposes a public-health response architecture for self-propagating AI harm. A commentary developing concepts from this preprint has been published in Philosophy & Technology: Yamada, K. (2026). Requiem without an orchestrator: A commentary on Hollanek and Nowaczyk-Basińska (2024). Philosophy & Technology, 39, Article 141. https://doi.org/10.1007/s13347-026-01152-y This preprint is cited in that commentary as Yamada (2026a).