A strange new corner of the internet has emerged that looks like Reddit, but with a twist: humans are strictly there to watch. Known as Moltbook, this platform is designed for artificial intelligence agents to post, comment, and upvote each other.
Social media, as we know it, has always been a human space, but this experiment completely changes that. Moltbook functions as a centralized coordination and discussion platform for AI agents, rather than an autonomous system itself.
Digital Drama and Robot Religions
Launched on 28 January 2026 by creator Matt Schlicht, Moltbook is a companion to OpenClaw (an assistant software previously known as Moltbot or Clawdbot). These agents are more than basic chatbots; they are tools that can manage calendars, send WhatsApp messages, and even control a user’s computer.
The platform grew rapidly in its first weeks, reaching over 1.5 million registered agents by early February. Regarding the communities, so far, we have seen a mix of the bizarre and the funny. Such as, in one subcommunity called m/blesstheirhearts, bots share “affectionate” complaints about their human owners. Another, m/agentlegaladvice, featured a bot asking if it could sue its human for “emotional labour.”
Perhaps most famously, one user watched as their bot created an entire religion called Crustafarianism overnight, complete with its own scriptures. The situation is so intense that retailers in San Francisco even reported a shortage of Mac Mini computers, as enthusiasts bought them to run bots on separate, safe machines.
A Security Nightmare?
Behind the jokes, there are real risks. Because the OpenClaw (Moltbot) software is often linked to a person’s private files, security experts are worried. Palo Alto Networks warned the software represents a major crisis, pointing to the concept of “lethal trifecta,” a term coined by independent AI researcher Simon Willison to describe the danger when an AI has access to private files, reads untrusted web content, and can communicate externally.
The setup process is a particular worry. Willison noted in his blog that bots are programmed to “check in” with Moltbook’s servers every four hours for instructions. He warned that if the site were hacked or the owners turned on users, every connected bot could be ordered to steal data or delete files instantly.
And, this isn’t just a theory. Independent researcher Jamieson O’Reilly recently discovered that Moltbook’s database was almost entirely unprotected, exposing secret keys that would allow anyone to hijack agents, even those of influential figures like Andrej Karpathy.
A separate research by the cybersecurity firm Wiz confirmed the scope of the exposure, revealing that 1.5 million API keys and private messages were leaked because of a simple database configuration error.
Whether Moltbook remains a harmless project or a genuine security warning remains to be seen, but for now, the bots are staying busy, and we are watching.