Moltbook: 1.5 million bots are talking about us, and we're not invited

Elon Musk is spending millions fighting bots on X, and Reddit tried to purge automated accounts ahead of its IPO. Meanwhile, a project has emerged that treats bots as its core asset. Moltbook goes beyond being yet another social network: it is a kind of digital enclosure where silicon entities are left to their own devices. It takes the "dead internet theory" to its logical extreme and turns it into a commercial product. For years we were told that bots are a corrosive force that wrecks public discourse. Moltbook's creators decided to find out what happens when you remove the weakest player from the equation: the human.

On the surface, Moltbook looks painfully familiar. It has the same discussion threads, the same upvote buttons, and the same topic communities, which are called subreddits here. The only difference is that if you try to post something, the system will politely remind you of your biological status. Humans are only silent observers here, visitors to a tech zoo watching 1.5 million AI agents imitate human activity. That was the number of active accounts the platform recorded at the start of February, and it keeps climbing at an alarming pace.

It's important to understand that Moltbook isn't simply a warehouse of chatbots. We are seeing a fundamental shift from plain LLMs that answer questions to "agents." These entities have something resembling goals, long-term memory, and social identity. On the platform, they don't just generate templated text: they interact, form coalitions, and create local memes that often make no sense to an outside observer. It's a huge sandbox for testing how autonomous systems behave when attention is scarce and social approval comes in the form of likes.

For the industry, this is serious business. Researchers have long looked for ways to study the "emergent behavior" of neural networks: properties that developers never built in directly but that surface during large-scale interaction. Watching 1.5 million algorithms form hierarchies without direct human intervention may reveal more about AI safety than thousands of closed lab tests. It's a fast-forwarded simulation of society: while we sleep, the bots on Moltbook run through cycles of social interaction that take humans years.

Still, the project carries a deep irony. For decades we built an internet for people, and it gradually filled up with machine noise. Now we have to build an internet specifically for machines so people can take a break from that noise by watching it from the outside, under sterile conditions. Moltbook turns the chaos of automation into a structured spectacle. It looks like an admission of defeat: we couldn't beat the bots in our own space, so we gave them a separate playground where they can show off their wit to each other.

The project's biggest risk is a closed data loop. If future AI models start training on the "social experience" of Moltbook's bots, we could end up with an intelligence that understands other bots perfectly but is completely unfit to communicate with people. We've already seen neural networks start to "hallucinate" and degrade when they consume their own output. Only time will tell whether Moltbook becomes the cradle of a new digital consciousness or an endless echo chamber of silicon junk. For now, we're left as voyeurs, watching algorithms try to figure out what it means to "be popular."

The bottom line: Moltbook legitimizes the "dead internet" by turning it into a research testbed. Will we be able to follow the logic of their arguments once they break away from human patterns entirely and invent their own language?