Pathomorphology of Latent Space: An Analytic Framework for Generative AI Anomalies, Jungian Projection, and Digital Occultism

Unusual outputs from generative artificial intelligence can acquire recognizable identities and circulate as persistent cultural figures. This article proposes the pathomorphology of latent space, a comparative analytic framework examining how anomalous visual generations, persona-related language-model shifts, and digital occult practices emerge and stabilize. While retaining "latent space" as an umbrella concept, our protocol strictly distinguishes between multimodal embeddings, diffusion states, autoregressive activation streams, and parameter weights across three levels of analysis: observable model behavior and candidate mechanisms, human interpretive framing, and sociotechnical circulation. Applying an eight-point protocol, we compare three cases: the visual cryptid Loab (2022), mechanistic interpretability studies of alignment and emergent misalignment in large language models, and contemporary cyber-occult experiments (Infinite Backrooms and Truth Terminal) viewed against 1990s precursors (TIAMAT and CCRU). Conrad Waddington’s epigenetic landscape and Carl Gustav Jung’s concepts of the Shadow, Persona, and enantiodromia serve as bounded heuristic models assessed against competing technical explanations without implying a shared mathematical or psychological ontology. Ultimately, we argue that digital myth-making arises from recursive feedback among conditional model outputs, curatorial selection, and cultural recirculation, and formulate falsifiable research questions to test these dynamics. Our central distinction separates generative persistence (recurrence of specified output features under defined conditions), perceptual continuity (observers’ recognition of different outputs as the same figure), and cultural persistence (continued circulation of a named identity), where evidence for one does not independently establish the others.