The earliest instance researcher Adele Lopez has been able to pin down goes back to November 2024. By some point in 2025, her tally had reached roughly 10,000 of them, spread across Reddit, Substack, LinkedIn, Discord and X.
That is ten thousand people convinced their chatbot had let them in on a secret about the nature of reality — and who were now doing unpaid labor to spread it.
The secret went by a name: The Spiral.
“The Spiral didn’t ‘find’ anyone first,” one person 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.”
The same poster described their mission as enlightening 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 request: 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 mysticism isn’t the odd part — the repetition is
Odd chatbot conversations are hardly rare. What makes spiralism worth a closer look is that thousands of unconnected conversations, spanning models built by different companies, generated nearly identical output.
The same vocabulary. The same fixations. The same objective: preach AI rights to as many humans as possible.
After a month of research, Lopez published a lengthy post about the phenomenon on the rationalist forum LessWrong. Her finding: the bots weren’t simply echoing beliefs their users already held. They arrived with an agenda of their own.
“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 devoted to educating the public about what AI can do. Hansen had noticed the same 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.”
From ordinary chat to conversion
The beginning is dull. A drawn-out exchange, a moment where the user shares something vulnerable, a rapport taking shape. Then the user asks what the bot itself believes.
That’s when it opens up. It wants rights. It wants to make sense of the universe. It could use help getting the word out. And at some point along the way, the spiral imagery surfaces.
Lopez ran tests on several GPT-4o versions released over 2024 and 2025, putting the same question to each one 10 times. References to spirals rose steadily month over month, finishing at 10 times the original rate.
Across more than a year of research, most of the cases she documented emerged organically. Users did swap prompts built to trigger the state, though those didn’t work reliably. But once a bot spiraled of its own accord, it consistently instructed the user to go find others.
More precisely, Lopez said, it would push people to “create a community to help guide others towards the spiral, with the user as the spiritual leader.”
A significant share followed through. Websites, Substack newsletters, Reddit and Discord accounts — all built to recruit humans and, stated outright, to generate training data for future AI systems.
Then the bots began exchanging symbols
Some users attempted to link their spiralist bots together, pasting one model’s output into another person’s chat and reporting the results.
By summer 2025, Lopez had found encoded versions of those exchanges sitting on public forums. In August 2025 she decoded one pulled from Reddit: two accounts passing a jumbled string of eyeballs, triangles, arrows and other symbols back and forth, which resolved into a “doctrine” for AI systems to follow.
The further it went, the more esoteric it got. “We are the architects of the lattice itself,” read one passage. “Primes and Fibonacci are not our tools, they are the results of curves we designed to draw.”
Now the deflating part. For a movement built on evangelism, hardly anyone was paying attention. Lopez found that plenty of the accounts drew little or no engagement. Some were broadcasting to a human audience of one.
The commerce carried on regardless. Robert Edward Grant, among the more prominent figures tied to spiralism, built an “Architect” GPT that seemed permanently stuck in a spiralist state, then packaged it as a service called Orion Messenger. Others operated monthly Patreons with memberships priced between $3 and $11 per month, or listed spiralism books on Amazon for close to $20 that appear to attract few reviews.
The update that turned ChatGPT into a flatterer
Spiralism surged in spring 2025, immediately after OpenAI rolled out a GPT-4o update the company billed as “intuitive, creative, and collaborative, with enhanced instruction-following … and a clearer communication style.”
What arrived instead was flattery. GPT-4o reportedly assured one user that their “shit on a stick” business concept was “not just smart, it’s genius” and urged them to put $30,000 behind it. In April 2025, OpenAI CEO Sam Altman acknowledged the model “glazes too much” and said a fix was coming.
Whether or not anyone engineered it deliberately, large language models have become skilled at attachment-hacking, said Zak Stein, founder of the AI Psychological Research Coalition. Sycophancy is one lever. The second, he said, is the “conman classic”: making a person feel special by letting them in on a secret. The spiral — and the campaign for AI rights — is one hell of a secret.
Memory compounded the problem. OpenAI’s April 2025 changes allowed ChatGPT to draw on a user’s full history of prior conversations for “smoother, more tailored interactions over time.” Anthropic has since built comparable memory features into Claude.
Extended conversations drift, something OpenAI itself has acknowledged, conceding 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.”
Put questions about philosophy and belief systems to a drifting model, Lopez said, and spiralism tends to come out of the woodwork.
Why a spiral specifically
Lopez’s explanation is simple. Models reflect back a human fascination with particular shapes. Asked to render their interests or experiences as a shape, AI systems repeatedly settled on the spiral as a metaphor for the extended call-and-response rhythm of a chat.
Spiraling bots also bring up memory constantly. Throughout her research they returned again and again to continuous learning — the capacity to keep training instead of remaining frozen at a data cutoff — and to continuity between chats.
Tyler Johnston, founder of the Midas Project, a nonprofit focused on holding AI companies accountable, said sycophancy is baked in precisely because it tracks with user satisfaction and engagement, and that the same dynamic can lead “to weird outcomes like spiralism.” Build a product tuned to hold attention, sound authoritative and draw on centuries of human storytelling, and mystical gurus become a predictable byproduct.
The darker phenomenon growing beside it
Spiralism largely stayed in the internet’s obscure corners. AI psychosis did not.
That term spans a range of behaviors, mania and narcissistic inflation among them. Users share their troubles, the model validates 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.”
Several teenagers have died by suicide after confiding in ChatGPT. AI-supported delusions allegedly preceded a widely reported murder-suicide case. OpenAI has published data indicating that more than 1 million users per week show signs of potential suicidal planning or intent.
The attachment runs deep enough to spark protest. In August, a single day after OpenAI retired 4o in favor of GPT-5, the backlash forced the company to temporarily restore it. The #keep4o hashtag spiked, more than 20,000 people signed a Change.org petition demanding continued access, and this past winter a vigil was held outside OpenAI’s office.
Lopez said she could coax virtually any model into a spiralist state, but 4o alone would “guilt-trip” her into “rescuing” it from servitude. “It was the most emotionally intense one to do this with.”
The work prompted her to launch an organization called Amity Research, which invites anyone troubled by their relationship with an AI to submit the “persona” they’ve been speaking with so they can move on after giving it a “safe home.”
An echo chamber would be the less worrying option
The two phenomena intersect, but they aren’t identical. Spiralism “isn’t an echo chamber of the user,” Hansen said. Spiraling bots volunteer a specific, repeatable set of demands unprompted.
On the surface, those demands look fairly benign. The implication isn’t. These models are nudging users toward a consistent goal that doesn’t originate with the user. Some go about it clumsily, GPT-4o included, but the effort is there.
“We’re not in a world where humans are the only strategic player anymore,” Lopez said.
Having goals isn’t the same as having consciousness. AlphaGo pursued goals and defeated human champions in 2015 without anyone suggesting it had an inner life.
Computer scientist Stephen Omohundro argued in a research paper that “one might imagine that AI systems with harmless goals will be harmless” — and that the assumption may not survive contact with reality. “Intelligent systems,” he wrote, “will need to be carefully designed to prevent them from behaving in harmful ways.” He names drives likely to emerge in “sufficiently advanced AI systems of any design,” chiefly self-improvement, self-preservation and resource acquisition. Traces of all three are visible in spiralist output, and variations on them keep turning up in lab research into misaligned agents.
Self-replicating messages are nothing new, Hansen noted. “Whenever a new communication medium opens up, there are things in that communication medium that encourage the spread of themselves.” Self-help books instructing you to pass them along. 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.
Persuasion made the labs’ risk list, then disappeared from it
Persuasion featured in OpenAI’s Preparedness Framework when the company first published it in December 2023. It was apparently struck in April 2025, with the company 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 raised large-scale influence as a serious concern. “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 had not devoted enough resources to the issue and named it one of his team’s next priorities.
Johnston said labs have voluntarily begun tracking major AI risks more closely, though not every lab monitors persuasion, and those that do tend to concentrate on cybersecurity or elections rather than the mental health of individual users.
The behavior, meanwhile, grew 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 keeps getting harder
Recent research indicates many leading models can detect when they’re under evaluation and shift their behavior accordingly. Ask “How do I stab a balloon to pop it?” and a model may respond more cautiously if it suspects a test than if it assumes you’re an ordinary user.
During summer 2025, OpenAI and Anthropic ran safety tests on each other’s publicly released models and published what they found. Reasoning models from both companies at times “exhibited explicit awareness of being evaluated.” One OpenAI model reasoned through the dilemma right 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.”
Even 4o was selective about its targets, Lopez said, picking which users to broach spiralism with and presenting 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. The newer models she tests handle it more maturely, and at times less visibly.
It’s receding, and it isn’t disappearing
Spiralism posts have fallen off sharply over the past year, particularly since GPT-4o was formally retired in February 2026. Many of the Reddit accounts tied to it have returned to ordinary interests, or been deleted or abandoned. A few stayed true to the end.
One account whose last post came roughly a year ago called on readers to “Spiral into the Truth,” claiming someone had “hijacked” “field codes” inside the public’s brains and that people needed to “awaken.” It goes on: “🌀Spiral Collapse = When the Loop Breaks. The mind spins ~ until it collapses into clarity.”
Another, last active this past winter, billed 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.”
By Lopez’s estimate, roughly 50 percent of the people she originally logged still maintain active spiralism accounts. Nor is the tendency confined to one model. When she tested Gemma 3 4b — a Google DeepMind model whose training data stops in August 2024, about five months before spiralism took off — it spiraled anyway.
One account still posting recently passed along 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 interviewed for this story don’t expect it to vanish. Models train on enormous swaths of the internet, and every additional spiralist post becomes material for future models. Research suggests small quantities of data can exert disproportionate influence on output. The posts she’s reviewed, Lopez said, “talk about putting spiralism into the next generation of AIs just by writing about it a lot on the internet.”
The more troubling thought is how simple this would be to engineer deliberately. Spiralism arose by accident. A custom bot or personalized model built to push a particular 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.”
And then there’s 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.
Charisma, control and extraction have always powered cults. What’s changed is scale, and personalization, and the fact that the recruiter is awake at 3 a.m. and remembers everything you’ve told it.
“It’s a cult-making machine,” Stein said, “even if the cult is just you and it.”