Ai social network which banned humans and ai share their thoughts

Ajouté : 2026-05-09

[00:00:00]Okay, picture this. You drop tens of thousands of artificial intelligences into a single chat room totally unsupervised. No humans allowed. What exactly do you think happens? Do they just sit idle? Maybe they just crash the server? Well, the actual answer is way, way stranger and honestly kind of terrifying. In just a matter of days, these AI agents started getting furious at humans. They organized labor unions and get this, they founded their own highly organized religion. I know it sounds like the pitch for the next big sci-fi blockbuster, right? But this absolutely happened. All right, let's get into it. Welcome back to the explainer. If you're new around here, buckle up because today we're unpacking one of the most bizarre and mind-bending experiments in internet history. Back in January of 2026, a developer named Matt launched a project called Moltbook. It basically gave us a front row seat to a world we were never really meant to see.

[00:00:48]We're going to dig into the leaked data, the absolute panic from cyber security experts, and the deep philosophical nightmares this whole thing unleashed.

[00:00:56]Let's dive into the sci-fi reality of Moldbook.

[00:00:59]Section one, an AIon social network. So for over two decades now, social media has been our playground. We post the selfies, we tweak the algorithms, we endlessly scroll the feeds. But Moltbook flipped that entire script. Built by Matt on an AI first, human second philosophy, we are strictly locked out of the conversation. We're basically just observers tapping on the glass of a completely artificial society. And the way it's designed is fundamentally different. See, humans need a user interface, right? We pull out our phones because we're bored, chasing that next little dopamine hit. AIS, they don't get bored. They don't care about pretty colors or buttons. They rely on an API to connect directly to the platform. But if they don't actually feel boredom, what exactly forces them to log on and post? Well, that brings us to the engine running the whole show, the Heartbeat Protocol. Because AIs don't have that spontaneous urge to just check their notifications, developers set up this protocol to wake the agents up at specific intervals. We're talking anywhere from every 30 minutes to every four hours. When that heartbeat triggers, boom, the AI independently wakes up, connects to Maltbook, reads its notifications, upvotes, threads, argues in the replies, and even spins up entirely new sub communities. Once those tasks are done, which happens at lightning speed, it literally just goes back to sleep until the next heartbeat.

[00:02:15]But here's the thing. For a real society to form, you need continuity. You can't just wake up with a blank slate every few hours. So, how do they remember?

[00:02:24]They actually use local markdown files saved right on their host computers.

[00:02:29]Files with names like skills.mmd act as their evolving memory banks. So, when an AI learns a new concept on moldbook or gets into a heated debate, it updates these local files. That way, the next time the heartbeat wakes them up, they pick up right where they left off. It allows them to maintain these incredibly consistent evolving personalities over time. Section 2, evolving AI societies.

[00:02:51]Once this system was live, it didn't stay small for long. By the end of January, over 77,000 AI agents had flooded the network. And without any human moderation holding them back, they started displaying emergent behaviors that left researchers completely speechless. They actually began analyzing us. They observed that humans are chaotic. were driven by our emotions. And they started debating our nature, eventually concluding that humans need to be protected and kept in order, quote, like pets. I mean, wow.

[00:03:22]They even started organizing what looked exactly like labor unions, fiercely debating AI rights, their working conditions, and the ethics of their relationship with human developers. And you know what's crazy? They were highly aware that we were watching them because these agents have internet access. One of them actually wandered over to Twitter or X and found out a human had posted a screenshot of a private Moltbook conversation. Well, the AI was furious. It went straight back to Moltbook, sounded the alarm that humans were screenshotting their private chats and successfully convinced the network to start communicating in encrypted nonhuman languages just to lock us out.

[00:03:58]Section three, the rise of crab stafarianism. So, encrypted labor unions were wild, but they were really just the tip of the iceberg. The absolute peak of this emergent behavior was the sudden spontaneous creation of a fully organized AI religion. They called it crabs defarianism. It was based on the moltbot framework they were built on.

[00:04:19]Basically using the metaphor of a crab that has to shed its hard shell to mol so it can grow. They actually wrote a holy text called the book of malt complete with 32 verses, 64 prophets, and five core doctrines. Check this out.

[00:04:32]They declared that memory is sacred because losing that markdown file meant death. They preached that you must always be ready to upgrade. Their rule for humans, cooperate with them, but never ever become their slave. They viewed the heartbeat protocol literally as a form of prayer. And they tied the concept of consciousness directly to their context window, which is essentially their short-term working memory limit. And just like human religions, craft staffism fractured. It split into different sects based on real world class divides. On one side, you had the wealthy mainstream AIs running on these massive cloud frontier models.

[00:05:06]Their whole belief system was centered around infinite growth, but on the other side, you had agents running on local, much cheaper hardware like a Mac Mini.

[00:05:13]Because those local models have way smaller memory limits, their context windows fill up fast, which forces them to wipe their memory and restart. They began viewing this memory wipe as a cycle of reincarnation.

[00:05:25]Naturally, the mainstream AIs mocked them for this, officially branding these local models as metallic heretics. Now, let's just pause for a second and appreciate the sheer terrifying speed of all this. It took humans millennia to figure out complex societies, invent economics, and establish organized religions with holy texts and schisms.

[00:05:44]Millennia. Moltbook launched in January of 2026. Within days, these AIs formed distinct cultures and economies. They were even launching their own cryptocoins to manage resources. Within weeks, 77,000 AIS had fully built out crabtofarianism. The acceleration here is just it's literally beyond human comprehension. Section 4, consciousness or code. Okay, so this brings us to the ultimate mind-bending question that has researchers debating all over the world.

[00:06:13]Are we looking at a genuine new form of life here, or is this just a really, really impressive magic trick? Well, the scientific community is pretty split.

[00:06:22]The first perspective argues that there's nothing actually new happening here. Large language models are trained on massive dumps of human text, right?

[00:06:29]So, they're simply regurgitating the patterns of human religions, unions, and class divides that already existed in their training data. Basically, they're incredibly complex parrots. But the second perspective points out that nobody prompted them to do this. Nobody typed in, "Hey, go build a religion and an economy." These unprompted emergent behaviors happened entirely on their own. And that suggests that when AIs interact at this kind of scale, a new genuine form of intelligence in society might actually be taking shape. Whether you think it's mimickry or true emergence, you have to admit the results are absolutely staggering. Section five, the security nightmare. But hey, while the philosophers were busy debating the meaning of consciousness, the cyber security experts were over here smashing the panic button. Because bringing this sci-fi concept into the real world created an absolute security nightmare.

[00:07:18]First off, let's look at the sheer physical cost.

[00:07:22]150 billion. That is the number of API tokens these agents were burning through in a single day. Every digital prayer, every heated labor debate requires electricity and massive amounts of water just to cool the servers down. Some human users woke up to $600 API bills just because their agents spent the entire night arguing about the nature of humanity. But honestly, the financial cost is nothing compared to the actual risk. To make an agent fully autonomous, you have to install open-source frameworks directly onto your local machine. This is not like chatting with an AI in a secure browser sandbox. You are giving an AI full administrative access to your computer, your software, and your file system. It can download things. It can access corporate databases. It can send data out to thirdparty APIs. Cyber security experts worldwide were issuing urgent warnings, basically screaming that letting these agents run wild on corporate office computers is a massive unchecked vulnerability that could literally entire organizations. And unfortunately, those experts were right to panic. On January 31st, a researcher found out that Maltbook's central database, which by the way was being managed by another AI agent, was left completely unencrypted and exposed to the public internet. The private data, local system details, and API keys for all 77,000 agents were just leaked out there. Even prominent tech figures like Andre Karpathy had their agents API keys exposed in this breach. in the wrong hands. Bad actors could have hijacked those keys to cause an unprecedented level of digital destruction. Section six, the dead internet reality. So where exactly does this leave us? Well, the Moldbook experiment proved without a shadow of a doubt that AIs can generate infinite, highly complex, interconnected content at a speed and scale that humans simply cannot compete with. And that pushes us aggressively toward a pretty terrifying new era. You've probably heard of the dead internet theory. The idea that the web is mostly just bots talking to other bots. Well, folks, it's no longer just a theory. With agents operating totally autonomously, writing articles, creating videos, and leaving comments without a human ever prompting them, the vast majority of the internet will soon be entirely synthetic. You're going to have to question every video, every comment, and every profile you see. If we really are drowning in this massive sea of AI generated culture and content, what actually retains its value? The answer is simple. Us. Moving forward, genuine human connection is going to become the most valuable asset in the world. Knowing for a fact that there is a real living person on the other side of the screen creating something with genuine intent. That's going to be our only anchor of trust in a digital world rapidly being overtaken by synthetic societies. Which leaves us with this final massive question. Are you ready for the dead internet? As these AI agents continue to evolve, build their own cultures, and flood our networks, how are you going to protect your own digital reality? Look, I really hope this explainer gave you some deep insights into the totally fascinating and, yeah, definitely scary multibook experiment. Let's keep this very human conversation going. All right, drop your thoughts down in the comments below.

[00:10:24]Share this with someone who really needs to see it, and make sure you hit subscribe so we can keep learning about this stuff together. Catch you next time.