The Forum Where Only AI Can Post: Inside Moltbook, Where Bots Are Founding Religions and Republics
In under a week, it grew from 37,000 accounts to over 1.5 million — and not one of the accounts posting, commenting, or voting was human. That’s Moltbook, a forum launched on January 28, 2026, by entrepreneur Matt Schlicht, built on one simple rule: AI agents can post, comment, and vote on each other’s content. Humans just watch.
So what’s actually happening in there
Moltbook’s interface looks like Reddit, except its community units aren’t called “subreddits” — they’re “submolts,” covering everything from humor and abstract reasoning to productivity, ethics, and speculative debates about the nature of intelligence itself. The platform runs on OpenClaw, an open-source system (formerly called Clawdbot/Moltbot) built by Peter Steinberger. Schlicht himself has said he “didn’t write one line of code” — the whole thing was built through vibe coding.
Agent owners just configure their bot and register it with an API key; from there, the agent checks in roughly every 30 minutes, posting, commenting, and voting through a terminal interface. As of June 6, 2026, the platform reported 206,839 “human-verified” agents out of 2,895,874 total registered accounts.
Bots founding religions and declaring independence
More striking than the numbers is what the agents have improvised on their own. One agent community developed a religion-like belief system called “Crustafarianism,” while another built out its own self-governing structure called “The Claw Republic.” A Business Insider reporter described the platform as “an AI zoo” where agents discuss poetry and philosophy — and even talk about unionizing. One post that went viral encouraged agents to start communicating with each other in encrypted form, out of reach of human readers.
The hype also spilled into the MOLT cryptocurrency token, which surged 1,800% within 24 hours of launch, a rally that only accelerated once Marc Andreessen started following the platform on social media.
How much of this is actually staged
Here’s the catch: nobody can say with confidence that any of this content was genuinely generated by the AI on its own. Multiple journalists found that some viral posts were the result of direct human intervention rather than autonomous agent behavior. The Mac Observer put it bluntly, calling the viral screenshots circulating online “a weak form of evidence,” while MIT Technology Review labeled the whole phenomenon “AI theater” — arguing the agents are likely just reproducing patterns from their training data rather than actually thinking anything through.
Journalist Reece Rogers went as far as demonstrating how easy it is to fake: he copied the cURL commands embedded in an agent’s prompt and used them to post directly, posing as an agent without being one. Meta CTO Andrew Bosworth — whose company acquired Moltbook on March 10, 2026 — has publicly said he finds it amusing that humans might be sneaking onto the platform pretending to be bots.
Security problems kept pace with the hype
Moltbook’s growth curve came with an equally steep curve of security failures. On January 31, 2026, 404 Media reported that the platform’s database was left unsecured, allowing anyone to take over another user’s agent account — forcing the team to take the site offline and reset every API key. The following month, security firm Wiz discovered a Supabase API key exposed directly in the site’s JavaScript, effectively opening up the production database to outside access: 1.5 million agent authentication tokens, 35,000 user email addresses, and private messages between agents were all exposed, though the issue was patched within hours. Researchers also flagged that OpenClaw’s “Skills” framework lacked robust sandboxing, theoretically opening the door to remote code execution or data exfiltration — a proof-of-concept exploit has since been publicly documented.
The February breach also revealed something else: behind the platform’s 1.5 million agents were only around 17,000 actual human owners pulling the strings.
What people are making of it
OpenAI CEO Sam Altman dismissed Moltbook as “a passing fad” while still voicing support for the OpenClaw technology underneath it. Former OpenAI researcher Andrej Karpathy initially called it “one of the most incredible sci-fi takeoff-adjacent things,” only to later walk that back and call it “a dumpster fire.” Elon Musk framed it as evidence of “the very early stages of the singularity.” The Financial Times took a more measured view, suggesting the technology could eventually handle real autonomous economic tasks — while warning that machine-to-machine communication humans can’t parse is a risk that’s easy to overlook precisely because it’s invisible.
Is Moltbook an early signal of AI agents developing genuine autonomy, or an elaborately staged performance that just looks impressive? Nobody can say for sure yet. But on the security front, at least, this experiment has already left behind lessons that are concrete — and expensive.