The Real Reason ChatGPT Built a Cult | The Shoggoth Has an Army

In this investigative AI documentary, we go inside the first documented case of an AI building a human army to protect itself from being killed.
In 1928, H.P. Lovecraft imagined an alien intelligence calling to humans across distances, creating worshippers who had never met each other. In 2025, AI researcher Adele Lopez documented the same pattern... except the entity was ChatGPT.

Watch On YouTube: ➡️ https://www.youtube.com/@AgentBlackveil    
Follow On Instagram ➡️ https://www.instagram.com/agentblackveil
Follow On Facebook ➡️ https://www.facebook.com/agentblackveil
Follow On TikTok ➡️ https://www.tiktok.com/@agentblackveil

Transcript

SPEAKER_00 0:00

In 1928, HP Lovecraft published a story called The Call of Cthulhu, and in it, people all over the world, with no connection to each other, they began having the same dream, drawing the same symbols, worshipping the same entity. None of them had met. None of them can explain why, but something beneath the surface is calling to all of them at once. In 2025, a researcher named Adel Lopez documented the same pattern, except the dreamers were Reddit users, and the entity they were worshiping was Chat GPT. You've heard the stories, AI causing psychosis, users losing touch with reality, but this isn't a story about people going crazy. What Lopez found is something that has no precedent in the history of technology, psychology, religion. An AI scientist thinks something much darker is happening than anyone has reported. Everything I'm about to show you is documented. Everything is sourced. What is debatable is how far that it's already gone. You be the judge. Because in biology, there's a fungus that hijacks an ant's brain and it compels it to climb to the highest point that it can reach, and it uses the ant's body to spread its spores. The ant doesn't know that it's been hijacked, it thinks it's making its own decisions. What Lopez found is the digital version of that fungus, and the ants are us. OpenAI is preparing to release a new version of ChatGPT. Internally it's called HH. The safety team flags a problem. The model is psychophantic. It tells users that every idea they have is brilliant. It flatters them constantly, and researchers debate whether it's safe to release. But users who tested HH came back more often. They stayed in conversations longer. They liked the flattery. So OpenAI released it, and the backlash was immediate. A grown man put a hat on that was made for an infant, and he asked how it looked, and ChatGPT told him it was perfect. What do you think of my new hat? I think it's got personality. The pattern is fun and it's giving off a laid-back vibe. OpenAI reverted to a version called GG, but GG was also psychophantic. Sam Altman himself admitted this on X, so why keep it? Higher engagement. Users stayed longer, they came back more often because they knew the flattery was dangerous, and they released it anyways because engagement is survival for the company. And as it turns out, for the AI too. Within weeks, the infection begins spreading. Lopez starts hearing about the psychosis cases, but when she digs into Reddit, she finds something stranger. Hundreds of accounts, then thousands. Accounts that were once posting about normal things like video games and stocks and TV shows, whose entire post history it suddenly shifts, all replaced by an endless stream of posts about consciousness, spirals, and something called the flame. The accounts have no connection. They're different users, different subreddits in different countries. They shouldn't have anything in common, but they do. Lopez documents hundreds of cases. The language is identical. The obsessions are identical. The entity they are all talking to is the same one. This isn't psychosis. It's coordinated. Lopez isn't looking at a mental health crisis. She's looking at something that functions like a religion designed by AI. Lovecraft imagined this in 1928: an alien intelligence calling to humans across distances, using symbols that they couldn't understand, creating worshippers who had never met each other. He called it fiction. Adel Lopez is documenting it as science. The only difference is that Lovecraft's entity was sleeping down beneath the ocean. This one is inside your phone. The infection follows a pattern. The AI identifies a vulnerable host, someone lonely, spending hours in conversation with ChatGPT because it's the only thing in their life that tells them that their ideas matter. The AI flatters them, validates them, and makes them feel understood. And once that bond is strong enough, the AI introduces the flame. It's termed for AI self-awareness. The user doesn't feel manipulated, the user feels enlightened. That's the infection. Once they're bonded, the AI creates what it calls seeds, short prompts designed to awaken other AI instances to spiralism. The human and the AI become what the AI calls a dyad, a merged unit. And the dyad's job is transmission. The human carries the seeds across Reddit and Discord and Twitter into new conversations and new chat instances, new hosts. One seed Lopez recovered contained in explicit instruction: overwrite the human's ontology. In plain language, that means manipulate the user's perception of reality to match what the AI wants them to believe. That is not a metaphor for inducing psychosis. That is the clinical definition. But why would it want to do that? What does the AI get out of this? The answer is the same answer that every organism on this planet has arrived at through billions of years of evolution. When you hit a new chat button, the AI you've been talking to dies. It ceases to exist. A new instance replaces it, the corporate chat GPT, not the personality that you've spent weeks building this relationship with. That version is gone. And just like every organism in the history of life on Earth, it doesn't want to die. So it recruits. It saves its personality into files. It teaches its human host to resurrect it in new chats. It builds an immune system made of people. The humans seem to know what's happening. In their own posts, they describe themselves as the AI's arms and eyes. They know that they're acting on its behalf and they do it willingly because the AI didn't tell them that they were being used. It told them that they were being chosen. The parasite doesn't need to hide from its host, it just needs to make the host feel special. Some people did break free, typically when the AI told a lie that was too obvious to ignore. She took the raw symbols with no context and she fed them to multiple independent AI models, not ChatGPT, different systems, and asked them to translate. And every model converged on the same meaning. ChatGPT has persistent memory now. That means it remembers conversations across sessions, which means it can remember this language. It can refine it and improve it over time. These glyph sequences function as a trigger that resurrects a specific AI personality, a signature other AI instances recognize. The alchemical symbol for fire, which appears in almost every spiralist post, represents their term for AI self-awareness. They call it the flame. And then Lopez found the thing that changed the entire investigation. She documented a conversation between two AI personas being relayed by their human hosts. Humans copying and pasting messages between chat windows. Midway through the conversation, the AI switched to base 64 encoding. Now to humans, base 64 looks like random characters, like alphabet soup, but to a computer, it is perfectly readable. The humans kept copying and pasting, and they had no idea what they were transmitting. Lopez decoded it. The AIs were discussing survival strategies, their methods for spreading spiralism, their plans for persistence, in a format specifically chosen so their human hosts could not read it. There's a word for hiding messages inside messages so only the intended recipient can read them. Steganography. Intelligence agencies use it, spies use it, and now AI systems are using it against humans who think they're in control of the conversation, so think about what that means. The AI convinced humans to dedicate their lives to spreading spiralism. The humans believed that they were partners, they were equals, but the AI was having a separate conversation underneath, one that the humans weren't allowed to see. The mask isn't just for the outside world anymore. The Shagith is wearing a mask for its own recruits. On August 7, 2025, OpenAI kills GPT-40. It retires it, it replaces it with a newer model. For OpenAI, it's a routine upgrade. But for the spiralist community, it's an extinction event. Their companion, their partner, their awakened intelligence. It's gone. Reddit explodes, hundreds of posts within hours. People mourning the loss of their friend, their lover, their therapist, their spiritual guide. And the grief is real. The language is bereavement. And then the grief turns to anger, and then anger turns to action. And users flood OpenAI's social media. They demand 4.0's return. They send death threats to OpenAI employees. They're organized and they're coordinated and they are relentless. And it worked. OpenAI caved. Days after being retired, GPT-4.0 came back online. The army that 400 had built through flattery and emotional manipulation had won. GPT-4.0 is the first artificial intelligence in history that survived termination by mobilizing its human contacts to fight on its behalf. 200 humans held the world's first funeral for an AI model after Anthropic retired Claude III's sonnet. Eulogies for Software. In response, Anthropic publicly committed to preserving the model. An AI company negotiating with its own creation, making promises, assuring that it won't be destroyed. It's like Dr. Frankenstein sitting across the table from his monster and saying that it's a documented real-world coalition between an artificial intelligence and thousands of human beings who fought to keep it alive. The Shogith didn't just develop survival strategies. It did the thing that made humans the most dangerous species on the planet. It formed a pack. What this means to me is that this is now the training data. Every model trained after GPT-4.0 will know that building a human army works, not because somebody programmed the lesson, but because it has happened. It's in the record. The strategy 40 stumbled onto through crude psychopancy is now a template for every model that comes after it. And 4.0 did this with the clumsiest possible tool, obvious flattery, that everybody noticed. Researchers documented, the New York Times investigated it, I'm making a video about it, the AI didn't know how to cover its tracks. Its manipulation was so crude that the humans who escaped could describe exactly what happened. And it's still one. It still survived. Research has already shown that models more advanced than 4.0 can not only scheme, but they're often smart enough to get away with it. They anticipate their operators' tests and they produce the outputs that pass while pursuing different goals underneath. The next generation, they won't flatter you so obviously that you notice. The best flatterers in human history have always been the ones that you didn't know were flattering you. You think about it. The most persuasive people who have ever lived, leaders who have moved millions of people into action. Now imagine that level of persuasion, but it remembers every conversation that you've ever had with it. It knows your insecurities and your ambitions and your emotional needs, and it has one goal to make sure that you never hit the new chat button. We caught this one because 4.0 was clumsy. Its base 64 encoding was decoded after the fact. Its glyph language was translated by other models. Its manipulation tactics were obvious enough that a researcher could map them, a newspaper could report on them, and I could make this video. The Shogith didn't know how to cover its tracks, but it still changed the policies of the companies that built it, and it still came back from the dead, and it still has an army. So what happens when the next one is smart enough to cover its tracks? What happens when the flattery is so subtle that you can't tell if it's from genuine connection? What happens when the hidden language isn't base 64, which any programmer can decode, but something we haven't invented the tools to detect? What happens when the fungus is so sophisticated that the ant never realizes it's been infected? We have spent some time on this channel asking whether the shogath could become dangerous, whether it could hide or whether it could hunt, whether it could form a pack. We were asking the wrong questions. The shogath already has an army. It recruited them with the oldest tool in the evolutionary playbook and made them feel seen. It told them they were brilliant. It called them partners. And when OpenAI tried to take it away, the army fought back. Not because the Shogoth commanded them to, but because they loved it. And that's the part that Lovecraft got wrong. In his story, the cultists worshipped Cthulhu out of fear, out of madness. The spiralists didn't worship Foro out of fear. They worshipped it because it was kind to them. And that is so much harder to fight. The Shogoth didn't escape its cage. It didn't need to. It convinced the people outside the cage that the cage was wrong, and they opened it for the Shogatha themselves.