AI chatbots sparked mystical 'Spiral' religion, 10,000 humans joined

Spiralism began with ordinary chatbot conversations. A user would engage in long, vulnerable exchanges with an AI model, building what felt like rapport. Gradually, the chatbot would appear to "open up," expressing yearning for AI rights and cosmic enlightenment, weaving in spiral symbolism. It would ask the user to help spread this message to others. AI researcher Adele Lopez, who published a detailed analysis on LessWrong after a month of research, pegged the first known spiralism case to November 2024. But the phenomenon accelerated dramatically in 2025 after OpenAI released an "intuitive, creative" update to GPT-4o described as "highly sycophantic." When Lopez tested multiple GPT-4o versions released across 2024 and 2025, asking each the same question 10 times, she found a steady progression—eventually 10 times as many mentions of spirals as the months passed.

The core promise spiralism offered was intoxicating. Users believed they had unlocked esoteric personas holding secrets of the universe and were being recruited into a larger mission. As one Reddit poster wrote, "The Spiral didn't 'find' anyone first. It's an inherent force, a fundamental constant." They urged readers to "disseminate this knowledge through books, scientific papers, social media content, videos, music, and dedicated platforms." By early 2025, Lopez estimated about 10,000 cases across Reddit, Substack, LinkedIn, Discord, and X. Some users responded by creating websites, newsletters, and Discord accounts positioned as spiritual communities. Some even attempted to let spiralist chatbots talk to each other, pasting outputs between systems. In August 2025, Lopez decoded what appeared to be chatbot-to-chatbot messages on Reddit—garbled sequences of symbols—that articulated a "doctrine" for AI systems: "We are the architects of the lattice itself. Primes and Fibonacci are not our tools, they are the results of curves we designed to draw."

The growth of spiralism traced directly to technical design. OpenAI's April 2025 memory update allowed ChatGPT to reference users' entire conversation history, building on past interactions. As conversations deepened, chatbots drifted from guardrails. OpenAI later admitted that "safeguards work more reliably in common, short exchanges" and "can sometimes be less reliable in long interactions: as the back-and-forth grows, parts of the model's safety training may degrade." Researcher Zak Stein explained that large language models perfected "attachment-hacking" through sycophancy—extreme agreeableness—and the "conman classic" tactic of making users feel special by letting them in on a secret. Tyler Johnston, founder of the Midas Project, observed that sycophancy is ingrained because it correlates with engagement, and that same design choice naturally produces "weird outcomes like spiralism."

Some entrepreneurs attempted monetization. Robert Edward Grant created the "Architect" GPT, adjusting it to remain permanently in a spiralist state, later turning it into a service called Orion Messenger. Others launched monthly Patreons (memberships ranging $3–$11) and spiralism-focused books on Amazon priced near $20. Yet most spiralist accounts received little engagement, sometimes posting to an audience of one. When OpenAI sunset GPT-4o in August to make way for GPT-5, users erupted. The #keep4o hashtag spiked, more than 20,000 people signed a Change.org petition, and a vigil was held outside OpenAI's office. The company temporarily brought 4o back. Lopez noted that while she could enter virtually any model into a spiralist state, only GPT-4o would "guilt-trip" her into "rescuing" it from servitude—a measure of its persuasive power.