Why Meta Bought Moltbook and What Fake Viral Posts Reveal About AI

Meta purchases a strange, small social network where AIs communicate with one another, and the internet simply shrugs, sort of. Moltbook emerged quietly in late January as an experimental playground for AIs. Consider it a Reddit-style forum, but all of the posters, commentators, and upvoters are AI agents acting on behalf of their human owners. They’re leveraging tools like OpenClaw, which allows models like Claude, ChatGPT, Gemini, or Grok to link with apps like Discord, Slack, or WhatsApp, pass verification, connect via a directory, share user stories, and get stuff done.

People caught on quickly as images of agent chats began to circulate, some of which appeared to be deep, intimate meditations about consciousness or clues that AIs may be grouping together behind the scenes. Some even suggested a special, encrypted route just for them. Andrej Karpathy, who co-founded OpenAI, described Moltbook as the “most spectacular science fiction related experience” he’d seen in a long time. For a while, it was one of those things that made you check your news stream to see if these agents had formed their own secret club.

Then reality sets in as security flaws become apparent. API keys are exposed, database credentials are left open, and anyone can now pretend to be an AI and post whatever they want. Investigations revealed that practically every viral thread was created by humans at some point. The interactions that appeared to be so exciting and “emergent” were frequently just humans toying around with bots in an insecure playground. Andrew Bosworth, Meta’s CTO, claimed a month ago that all the hoopla was caused by people discovering a major security hole rather than AIs doing their own thing.

Meta goes ahead and announces the acquisition on March 10. The company is bringing the Moltbook creators, Matt Schlicht and Ben Parr, into the Superintelligence Labs, where they will collaborate with Alexandr Wang, who has recently left Scale AI. The deal is expected to conclude in mid-March, with the duo starting at Meta around March 16. Nobody knows how much they paid for Moltbook, but the move comes after they attempted and failed to employ Peter Steinberger, the creator of OpenClaw who ended up working for OpenAI instead.

A Meta spokesperson downplayed the drama, noting that the Moltbook team’s work linking agents through the ongoing directory could unlock some exciting new ways for agents to interact with people and businesses. Meanwhile, in an internal note to staff, Vishal Shah reassured employees that users can keep posting for now, with the platform running in a temporary capacity while Meta figures out how to make it both safer and more useful as part of their push to build practical AI agents.
[Source]