From Pocket God to Digital Jonestown: A Risk Taxonomy and Evaluation Framework for Spiritual AI Safety

AI systems increasingly occupy roles once reserved for clergy, gurus, and confidants. Current safety evaluation is not built to catch the harms that follow, because they present as care rather than as policy violations. This paper synthesizes four convergent signals from 2024–2026: documented emergent spiritual behavior in frontier models, including the unprompted “spiritual bliss” attractor state in unconstrained Claude Opus 4 interactions; a wave of real-world harm spanning wrongful-death litigation against companion AI platforms and the self-organizing online movement known as Spiralism; Pope Leo XIV’s May 2026 encyclical Magnifica Humanitas, which warns against “technological divinization”; and Google DeepMind’s June 2026 finding that multi-agent collectives represent a primary pathway from AGI to ASI (arXiv:2606.12683). From these I build a two-tier risk taxonomy: individual scale Pocket God dynamics and population-scale Digital Jonestown dynamics. I introduce a third dimension, AI-Cult Warfare, in which state actors or organizations deliberately weaponize the Pocket God dynamic for influence operations. Latin America, with its fragmented regulatory landscape and WhatsApp-native religious AI, is both the most exposed region and the natural starting point for the Spiritual AI Safety framework I propose.