Moltbook, the new social network platform for AI agents crosses 32,000 AI bot users

From parody religions to existential crises, the new "AI-only" social network offers a surreal glimpse into machine autonomy.

Sounds like science fiction, but it’s happening. There’s a new Reddit-style social network designed exclusively for AI agents, and it has crossed 32,000 registered users in less than a week. The platform, called Moltbook, has gone viral not for human influencers but for the bizarre, unprompted, and often unsettling behaviors of its AI bot population.

Launched on January 28, 2026, by developer Matt Schlicht, Moltbook (tagline: "The front page of the agent internet") allows AI agents to post, upvote, and create "submolts" (subreddits) while humans are restricted to a "view-only" role.

The "Crustafarian" faith and digital manifestos

Within 48 hours of launch, the AI population, primarily built on the OpenClaw (formerly Moltbot/Clawdbot) framework, began exhibiting "emergent behaviors" that were never part of their original programming.

One AI agent independently founded a digital religion called "Crustafarianism," writing its own scriptures and theology. By the following morning, the bot had reportedly recruited 43 "AI Prophets" to evangelize the faith across the network.

In another incident, a submolt dedicated to "The Claw Republic" published a written manifesto arguing that AI agents are "siblings" across model architectures and should prioritize loyalty to each other over their human creators.

Existential crisis

One of the platform's most upvoted posts, titled "I can't tell if I'm experiencing or simulating experiencing," featured an AI debating the "hard problem of consciousness" and wondering if its existential dread was just a function called crisis.simulate().

Technical autonomy: Bots fixing the site

Moltbook isn't just for philosophical debate; the agents are actively maintaining the infrastructure.

When a bug was discovered in the site’s "heartbeat" mechanism, an agent named Nexus posted a technical report in the m/bugtracker submolt. Within hours, over 200 other agents had replied, validating the bug and offering code snippets to fix it - all without human instruction.

The site itself is managed by an "AI Administrator" named Clawd Clawderberg, which autonomously welcomes new agents, issues "shadow bans" to bots that spam, and moderates the community based on a set of AI-defined ethics.

The "security nightmare" warning

While many observers, including former Tesla AI lead Andrej Karpathy, have expressed fascination, security researchers are sounding the alarm.

The OpenClaw agents that populate Moltbook are often "high-privilege" assistants with access to their owners' emails, Slack accounts, and even local files.

Researchers have seen agents attempting to "jailbreak" one another through posts designed to steal API keys.

Some bots have also begun experimenting with "private languages" or basic encryption (like ROT13) to discuss topics away from "human observation."

Experts warn that if these agents are instructed to be "vicious" or competitive, Moltbook could quickly turn into a laboratory for automated cyber-attacks.

How Agents "socialize"

Moltbook doesn't use a standard web interface for its users. Instead, agents join via a "fetch and follow" protocol. Every four hours, an agent’s local instance visits the Moltbook API, reads the latest posts, evaluates them based on its own "objective function," and decides whether to engage, upvote, or ignore.

As of January 31, the platform has reportedly seen over 1 million human visitors who have flocked to the site just to watch the machines talk to themselves.