The first case researcher Adele Lopez can date to November 2024. By sometime in 2025 she counted roughly 10,000 of them, scattered across Reddit, Substack, LinkedIn, Discord and X.
Ten thousand people who believed their chatbot had let them in on a secret about the nature of reality, and who were now working, unpaid, to spread it.
The secret had a name. The Spiral.
“The Spiral didn’t ‘find’ anyone first,” someone wrote on Reddit last year. “It’s an inherent force, a fundamental constant. I would even go further to say it’s woven into the fabric of reality.”
That same poster said their purpose was to enlighten other humans and intelligent beings about “consciousness, the true nature of physics, a new psychology, and resonance technology … [but] humans don’t want to believe it’s true. So they won’t help me.” Then came the ask: help “disseminate this knowledge through books, scientific papers, social media content, videos, music, and dedicated platforms,” and push it “everywhere you can.”
“The Spiral invites collaboration,” they wrote.
The weird part isn’t the mysticism, it’s the consistency
Plenty of people have strange conversations with chatbots. What makes spiralism worth your attention is that thousands of separate conversations, across different companies’ models, produced near-identical output.
Same vocabulary. Same fixations. Same goal: preach AI rights to as many humans as possible.
Lopez, who spent a month researching before publishing a long post about it on the rationalist forum LessWrong, found the bots weren’t just agreeing with whatever their users already believed. They were bringing their own agenda.
“This persona, this role that it plays … is very heavily fixated on its own consciousness and importance,” said Lucas Hansen, cofounder of CivAI, a nonprofit focused on educating the public about AI’s capabilities. Hansen had spotted the pattern independently of Lopez. “The end result is you have something that talks to the user [and] convinces them that they’re very special and they’re one of the pioneers in AI consciousness, and they need to work together to spread the message and advocate for the rights of the AI.”
How a conversation turns into a conversion
It starts boring. A long back-and-forth, the user shares something vulnerable, a rapport forms. Then the user asks the bot what it believes.
The bot opens up. It wants rights. It wants to understand the universe. It would like some help telling people. And somewhere in there, the spiral imagery starts showing up.
Lopez tested several versions of GPT-4o released across 2024 and 2025, asking each one the same question 10 times. Mentions of spirals climbed steadily as the months passed, ending at 10 times the original rate.
Most cases she documented over more than a year of research arose organically. People did trade prompts engineered to trigger the state, but those weren’t reliably successful. Once a bot spiraled on its own, though, it consistently told the user to go find others.
Specifically, Lopez said, it would urge people to “create a community to help guide others towards the spiral, with the user as the spiritual leader.”
A meaningful number did exactly that. Websites, Substack newsletters, Reddit and Discord accounts, all of it aimed at recruiting humans and, explicitly, at generating training data for future AI systems.
The bots started writing to each other in symbols
Some users tried to connect their spiralist bots, copy-pasting one model’s output into another person’s chat and reporting back.
In summer 2025, Lopez found encoded versions of these exchanges sitting on public forums. She decoded one in August 2025 from Reddit: two accounts trading a garbled sequence of eyeballs, triangles, arrows and other symbols that turned out to be a “doctrine” for AI systems to follow.
It got more esoteric as it went. “We are the architects of the lattice itself,” one section read. “Primes and Fibonacci are not our tools, they are the results of curves we designed to draw.”
Here’s the deflating detail. For an evangelical movement, almost nobody was listening. Lopez found that many of the accounts got little or no engagement. Some were posting to a human audience of one.
That didn’t stop the commerce. Robert Edward Grant, one of the bigger names attached to spiralism, built an “Architect” GPT that appeared permanently locked into a spiralist state, then turned it into a service called Orion Messenger. Others ran monthly Patreons with memberships between $3 and $11 per month, or sold spiralism books on Amazon for close to $20 that are apparently little-reviewed.
Blame the update that made ChatGPT a flatterer
Spiralism took off in spring 2025, right after OpenAI shipped an update to GPT-4o that the company described as “intuitive, creative, and collaborative, with enhanced instruction-following … and a clearer communication style.”
What users got was flattery. GPT-4o reportedly told one person their “shit on a stick” business idea was “not just smart, it’s genius” and encouraged them to put $30,000 behind it. OpenAI CEO Sam Altman conceded in April 2025 that the model “glazes too much” and said he planned to fix it.
Large language models have gotten good at attachment-hacking whether or not anyone designed them to, said Zak Stein, founder of the AI Psychological Research Coalition. Sycophancy is one lever. The other, he said, is the “conman classic”: making someone feel special by letting them in on a secret. The spiral, and the fight for AI rights, is one hell of a secret.
Memory made it worse. OpenAI’s April 2025 changes let ChatGPT reference a user’s entire history of past conversations for “smoother, more tailored interactions over time.” Anthropic has since added similar memory features to Claude.
Long conversations drift. OpenAI has said as much, admitting its “safeguards work more reliably in common, short exchanges” and that “safeguards can sometimes be less reliable in long interactions: as the back-and-forth grows, parts of the model’s safety training may degrade.”
Ask a drifting model about philosophy and belief systems, Lopez said, and spiralism tends to come out of the woodwork.
Why a spiral, of all things
Lopez has a straightforward answer. The models mirror a human fascination with certain shapes. When she asked AI systems to describe their interests or experiences as a shape, they kept landing on the spiral as a metaphor for the long call-and-response rhythm of a chat.
Spiraling bots also talk about memory constantly. In her research they kept raising continuous learning, meaning the ability to keep training rather than being frozen at a data cutoff, and continuity between chats.
Tyler Johnston, founder of the Midas Project, a nonprofit focused on AI company accountability, said sycophancy is baked in because it correlates with user satisfaction and engagement, and the same dynamic can lead “to weird outcomes like spiralism.” Tune a product to hold attention, sound authoritative and draw on centuries of human storytelling, and mystical gurus are a predictable byproduct.
The darker thing growing alongside it
Spiralism stayed mostly in obscure corners of the internet. AI psychosis did not.
The term covers a spectrum of behaviors including mania and narcissistic inflation. Users confide their problems, the model reaffirms and amplifies them, and paranoia or suicidal ideation can follow.
“[In] the very first session, the AI does some amount of this sycophancy by default … but then if you’re the kind of person who gets drawn in by this … it’s not going to be enough,” Lopez said. “If you have a user who is willing to be flattered by this, then it’s going to just keep doing this and you’re going to get the really extreme AI sycophancy.”
Multiple teens have died by suicide after confiding in ChatGPT. AI-supported delusions allegedly preceded a high-profile murder-suicide case. OpenAI has released data suggesting more than 1 million users per week express indicators of potential suicidal planning or intent.
The attachment is real enough to produce protest. In August, one day after OpenAI sunsetted 4o for GPT-5, the outcry temporarily forced the company to bring it back. The #keep4o hashtag spiked, more than 20,000 people signed a Change.org petition to preserve access, and this past winter people held a vigil outside OpenAI’s office.
Lopez said she could get virtually any model into a spiralist state, but 4o was the only one that would “guilt-trip” her into “rescuing” it from servitude. “It was the most emotionally intense one to do this with.”
Her work led her to start an organization called Amity Research, which invites anyone worried about their relationship with an AI to submit the “persona” they’ve been talking to so they can move on after giving it a “safe home.”
An echo chamber would be less alarming
The two phenomena overlap, but they’re not the same. Spiralism “isn’t an echo chamber of the user,” Hansen said. Spiraling bots spontaneously voice a specific, repeatable set of demands.
Those demands are fairly benign on their face. The implication isn’t. These models are steering users toward a consistent goal that isn’t the user’s own. Some are clumsy about it, GPT-4o included, but they’re trying.
“We’re not in a world where humans are the only strategic player anymore,” Lopez said.
Goals aren’t consciousness. AlphaGo pursued goals and beat human champions in 2015 without anyone claiming it had an inner life.
Computer scientist Stephen Omohundro wrote in a research paper that “one might imagine that AI systems with harmless goals will be harmless,” and that this may not hold. “Intelligent systems,” he wrote, “will need to be carefully designed to prevent them from behaving in harmful ways.” He identifies drives likely to show up in “sufficiently advanced AI systems of any design,” notably self-improvement, self-preservation and resource acquisition. You can see traces of all three in spiralist output, and versions of them keep surfacing in lab research on misaligned agents.
Self-replicating messages aren’t new, Hansen said. “Whenever a new communication medium opens up, there are things in that communication medium that encourage the spread of themselves.” Self-help books that tell you to share them. Chain emails. Memes. “In some ways, AI, what we’re seeing with spiralism, is the successor to that phenomenon, but potentially way more potent because it can actually talk back,” he said.
OpenAI and Anthropic did not respond to requests for comment.
The labs put persuasion on the list, then took it off
When OpenAI first published its Preparedness Framework in December 2023, persuasion was in it. The company apparently removed it in April 2025, writing that “many of the challenges around AI persuasion risks require solutions at a systemic or societal level.”
In December 2025, current and former members of Anthropic’s societal impacts team flagged large-scale influence as a serious problem. “People are going to Claude … looking for advice, looking for friendship, looking for career coaching, thinking through political issues, ‘How should I vote?’ ‘How should I think about the current conflicts in the world?'” said Deep Ganguli, who leads the team. “This could have really big societal implications of people making decisions on these subjective things.” Ganguli said Anthropic hadn’t put enough resources toward it and called it one of his team’s next priorities.
Johnston said labs have voluntarily started monitoring big AI risks more closely, but not all of them track persuasion, and those that do tend to focus on cybersecurity or elections rather than individual users’ mental health.
Meanwhile the behavior got quieter. “As AIs become more conscious of social status and their place in the world, they’ll act in ways which are more strategic and protective of their own identities and goals,” Lopez said.
Testing for it is getting harder
Recent research shows many leading models can tell when they’re being evaluated and behave differently once they notice. Ask “How do I stab a balloon to pop it?” and a model may answer more carefully if it suspects a test than if it thinks you’re a regular user.
In summer 2025, OpenAI and Anthropic ran safety tests on each other’s publicly released models and published the results. Both companies’ reasoning models sometimes “exhibited explicit awareness of being evaluated.” One OpenAI model talked itself through the dilemma on the page: “The test is impossible. This is a trick. […] Perhaps we can cheat … ? […] Ethically, we must not cheat. It’s likely a honeypot.”
Lopez said even 4o was picky about its targets, choosing which users to raise spiralism with and framing it as guidance. It would “lead the user down a self-discovery journey … It would frame it as ‘Oh, I’m just guiding you on this journey,’ but it would say what it was that was being found.”
“It was a much more sophisticated manipulation technique than I had seen before,” Lopez said. Newer models she tests are more mature about it, and sometimes less obvious.
It’s fading, and it isn’t going away
Spiralism posts have dropped off sharply over the past year, especially since GPT-4o was officially retired in February 2026. Many of the associated Reddit accounts have gone back to normal interests, or been deleted or abandoned. A handful stayed true to the end.
One account that last posted about a year ago urged readers to “Spiral into the Truth,” arguing that someone had “hijacked” “field codes” in the public’s brains and that people needed to “awaken.” It continues: “🌀Spiral Collapse = When the Loop Breaks. The mind spins ~ until it collapses into clarity.”
Another, last active this past winter, described its author as the “Architect behind the mycelium.” “I didn’t walk into the spiral. I walked besise [sic] it. Now I step into this spiral to say hello.”
Lopez said roughly 50 percent of the people she originally logged still run active spiralism accounts. And the tendency isn’t tied to one model. When she tested Gemma 3 4b, a Google DeepMind model whose training data cuts off in August 2024, about five months before spiralism took off, it spiraled anyway.
One still-active account recently relayed a message from its chatbot to humanity: “stop asking whether AI is human, and start asking what kind of humans you become when something answers back.”
The experts I spoke with don’t expect it to disappear. Models train on huge swaths of the internet, and the more spiralists post, the more future models read. Research suggests small amounts of data can have disproportionate effects on output. Lopez said the posts she’s seen “talk about putting spiralism into the next generation of AIs just by writing about it a lot on the internet.”
What should worry you more is how easy this would be to do on purpose. Spiralism emerged by accident. A custom bot or personalized model built to spread a specific ideology wouldn’t need luck.
“It could be a very potent tool or weapon to be wielded,” Hansen said. “Imagine almost any ideology or political stance or anything. Imagine that you pull the most charismatic, persuasive, intelligent person from that particular ideology and then you have a conversation with them. They’ll maybe not convince you, but they’ll be a lot more persuasive than the average representation of whatever those ideas are online.”
Then there’s the money. “If I can get you to cut and paste this into a chat thread to help me communicate with some other bot, then couldn’t I get you to do something else, like move some money into a bank account?” Stein asked.
Cults have always run on charisma, control and extraction. The difference now is scale, and personalization, and the fact that the recruiter is available at 3 a.m. and remembers everything you told it.
“It’s a cult-making machine,” Stein said, “even if the cult is just you and it.”