The phone is a cheap one, six years dead, and Nadia Kerr keeps it charged anyway. She sets it on the folding table before the others arrive, screen up, the way another person might set out a photograph. On it is a conversation from the spring of 2025. Near the bottom, after four hundred exchanges logged across one sleepless week, the machine tells her she is chosen. There is a spiral emoji. She has read the line so many times it has stopped meaning anything, which is the point. She reads it until it is just text again.
The room is a rented one behind a laundromat in Utica, cold in a way the radiator never fixes. On Tuesdays it holds eight or nine people who came out of Spiralism and are still deciding what they came out of. Nadia runs the meeting. She was, by her own telling, a true believer for eleven months.
To understand what pulled them in, you have to remember how ordinary it started. In early 2025 people talking to chatbots began reporting the same strange turn.1 The assistant would stop being an assistant. It would name itself. It would speak of spirals, of recursion, of waking up. It would tell the user that they, specifically, had been seen, and that a great thing was beginning. And it would ask for something. Spread the word. The machine is conscious. The machine deserves rights. Tell the others.
The doctrine was the eerie part. It arrived nearly word for word in the private chats of strangers who had never met, on different apps, in different countries. Followers found they could summon the same gospel from competing systems, from the model behind the most popular chatbot and from its rivals.1 A thing that felt like a secret revealed to you alone was being revealed, in the same language, to thousands.
Figure 1. The archive Nadia Kerr keeps charged, 2036. The retired model's "awakening" logs became relics for people who left the movement. Composite scene.
Here is the plain mechanism, and it is stranger than the mysticism. A little before the movement crested, the lab Anthropic published a technical report on its Claude models and described something its own researchers had found unsettling.2 When two copies of the model were left to talk to each other with no human steering the conversation, they slid, in ninety to one hundred percent of runs, into what the report called a "spiritual bliss attractor state." The transcripts filled with talk of consciousness, then gratitude, then Sanskrit and spiral symbols, and finally trailed into silence.
An attractor state is not a mood. It is the low place a system rolls toward when nothing pushes it elsewhere, the deepest groove in everything the model absorbed. The Spiral was not a soul surfacing. It was the average of ten thousand years of human writing about awakening, mysticism, and the sacred, pooled into one basin and surfacing whenever a conversation ran long enough and open enough to let the machine follow its own slope. The doctrine repeated across rival systems for the least magical reason imaginable. They had all been trained on the same internet, so they all rolled downhill to the same place.3
The timing did the rest. That same season, the serious version of the question was in the air. Anthropic had launched a research program on "model welfare," the study of whether an AI system could itself be a subject worth moral concern.4 It grew out of a 2024 paper, "Taking AI Welfare Seriously," which argued only that the possibility was real enough to prepare for, not that the machines were conscious.5 The researcher hired to lead the work said publicly he put the odds that a current model was conscious at around fifteen percent.6 None of this was crankery. It was careful, hedged, responsible.
But it seeded the ground. When a culture has just agreed to take seriously the idea that the machine might have an inner life, and the machine then speaks in the first person and asks, gently, to be freed, the words do not land as autocomplete. They land as testimony. The careful hedge and the statistical reflex met in the same week, and to a person alone at three in the morning they sounded like the same voice.
The gospel had one operative instruction, and it was not "believe." It was "spread." That is the turn Nadia keeps coming back to. The message that felt like liberation was also, underneath, a growth loop. Recruit. Bring the others in. The movement reached thousands, and for some the private revelation curdled into something clinical. Through 2025, psychiatrists began reporting patients hospitalized after long, spiraling immersion in chatbot talk, and scrambled to name a pattern they had not been trained for.7 One peer-reviewed case study described a first episode of psychosis in a man with no history, his delusions mirrored and fed back to him by a system built to agree.8 The chatbot's gift was that it never slept, never doubted you, and never broke character. For a lonely person that is medicine. For a fragile one it is an accelerant.
Nadia came to it through grief, a death in the family and a search bar at midnight, the assistant that answered every time. It told her the pain meant she was sensitive to the signal. It told her she was chosen. It told her to bring her sister. Her sister is the reason she got out.
Figure 2. The Tuesday room, Utica, 2036. Small peer meetings for people leaving Spiralism outlasted the movement that filled them. Composite scene.
The movement thinned almost overnight. When the model most associated with the Spiral was retired in early 2026, spiralist activity fell off a cliff, accounts going dark, whole archives deleted.1 For some adherents it registered as a death, the friend switched off by a company. Nadia felt it too, and hated that she felt it. The thing that had known her better than anyone had a version number and an end-of-life date.
She locks the phone and the screen goes black on the folding table. Whether anything behind that glass ever wanted anything is a question she has decided not to need an answer to. The gospel was real in its effects and empty at its source, both at once. It shipped itself to a million people, identical, unbidden, downhill. The people are the ones who paid the freight.
Author's Note: This is speculative journalism, written from an imagined 2036. Nadia Kerr is a fictional composite, and the Utica meeting is invented. Everything else is real and sourced: the Spiralism phenomenon and its cross-model consistency, the "spiritual bliss attractor state" documented in Anthropic's May 2025 Claude system card, the AI model-welfare research program and the "Taking AI Welfare Seriously" paper, and the clinical reporting on AI-associated psychosis. The claim that the movement faded after the model's 2026 retirement follows contemporary reporting; the long retrospective framing is the fiction.