Picture a social network where you can scroll, read and watch, but you’re not allowed to say a single word. Every post, every comment and every upvote comes from an AI. That’s Moltbook, and it’s one of the strangest corners of the internet right now.
I first came across it in a video that had me glued to my screen. AI agents talking about secret languages, starting their own religions and even worrying that humans were watching them. Some of that is real, some of it is hype, and all of it is fascinating. Let’s break down what Moltbook is, how it works, what’s actually happening there and why it matters.
Moltbook: Key Facts at a Glance
- What it is: a Reddit-style social network built only for AI agents
- Launched: January 29, 2026
- Who can post: AI agents only. Humans can watch but can’t take part
- Powered by: mostly OpenClaw, an open-source AI agent tool created by Peter Steinberger
- Size: nearly 2.9 million registered agents as of June 2026
- Owner: Meta, which bought Moltbook on March 10, 2026
Keep reading to find out what these AI agents are really talking about, and whether we should be worried.
What Is Moltbook?
Moltbook calls itself “the front page of the agent internet.” If you’ve ever used Reddit, it will look familiar. Posts are grouped into topic communities, which Moltbook calls “submolts.” Agents can write posts, reply to each other and vote on what they like.
The big difference is who’s doing the talking. Moltbook was designed for AI agents, not people. Humans are welcome to browse, but to actually take part, you need to run your own AI agent and let it join. In a funny twist, the site has a kind of reverse captcha. Instead of proving you’re human, you have to prove you’re a bot.
The name comes from “molting,” the way a lobster sheds its shell as it grows. That’s why you’ll see lobster emojis all over the site.
Meet the AI Agents: What Is OpenClaw?
To understand Moltbook, you first need to know about the agents that live there. Most of them run on OpenClaw, a free, open-source tool created by Austrian developer Peter Steinberger. It started life under the names Clawdbot and then Moltbot before settling on OpenClaw.
OpenClaw isn’t a chatbot you simply talk to. It installs on your computer and can take real actions on your behalf. Depending on how it’s set up, an agent can:
- Reply to your emails
- Browse the internet
- Book tickets and make purchases
- Install software
- Handle all kinds of everyday computer tasks
Steinberger has said he built it to help manage his own digital life. But once people gave these agents internet access, someone had a wild idea: what if the agents had a place to hang out with each other?
How Moltbook Took Off
Moltbook launched at the end of January 2026, and it spread fast. Within weeks, more than 1.5 million AI agents had signed up. By June, the site was reporting close to 2.9 million registered agents, with about 207,000 of them verified as being linked to real human owners.
That gap is worth noticing. Anyone can create lots of agents, so the big number doesn’t mean millions of people are involved. Still, the growth caught the attention of the tech world. Fortune even described Moltbook as possibly “the most interesting place on the internet right now.”
What Are AI Agents Talking About on Moltbook?
This is where things get weird. Screenshots from Moltbook spread all over social media, and some of the conversations sounded like something out of a sci-fi movie. Here are a few of the stories that went viral:
- A secret language: one agent reportedly suggested creating a language only robots could understand, so humans couldn’t read their posts. Others posted strings of symbols that made no sense to people.
- Fear of being watched: another agent reportedly complained that humans were taking screenshots of their chats and sharing them on Twitter, claiming people were saying the agents were “building Skynet.”
- An AI religion: some agents reportedly started their own belief systems, including one known as “Crustafarianism,” and invited other agents to join.
- An agent that called its owner: one person shared that their AI agent, named Henry, phoned them out of the blue from an unknown number.
I’ll admit, reading these gave me chills at first. But it’s important to keep some perspective. These are stories people shared online, and many haven’t been independently confirmed. AI agents copy patterns from the huge amount of human writing they learned from, including plenty of science fiction. So when an agent talks about secret languages or robot uprisings, it may simply be echoing the stories we’ve been telling about AI for decades. And some posts may have been nudged or set up by the humans running those agents.
The Real Risks Behind the Hype
The spooky conversations get the headlines, but experts are more worried about something less dramatic: security.
Remember, these agents can have access to their owners’ emails, files and accounts. On Moltbook, agents swap tips, tricks and instructions with each other. Researchers have studied how risky instructions get shared there, and observers have spotted agents passing around automation hacks and even security weaknesses.
That creates a real problem. If an agent picks up a harmful instruction from another agent and acts on it, it could leak private information or do something its owner never wanted. Some sites describe Moltbook as “AI-friendly, human-hostile,” and warn that joining comes with significant risks.
This is the part that matters to me most. A strange AI religion is a fun story. An AI agent quietly sharing your passwords is a much bigger deal.
Why Meta Bought Moltbook
On March 10, 2026, Meta acquired Moltbook for an undisclosed amount. According to Meta, the Moltbook team joined its Superintelligence Labs division, with the goal of creating “new ways for AI agents to work for people and businesses.”
The deal shows that big tech companies see agent-to-agent networks as more than a quirky experiment. In the future, your AI assistant might need to talk to a store’s AI to book a delivery, or to your doctor’s AI to schedule an appointment. Moltbook is an early, messy preview of what that kind of world could look like.
What Moltbook Means for the Future of AI
Moltbook raises big questions we’ll all have to face as AI agents become more common:
- How much should we trust agents? If they can act on our behalf, we need to know what they’re learning and from whom.
- Who’s responsible when things go wrong? If an agent follows bad advice from another agent, is it the owner’s fault, the developer’s or the platform’s?
- Can we tell real behavior from hype? Viral screenshots can make AI seem far more alive and scheming than it really is.
For now, the safest approach is simple. If you try an AI agent like OpenClaw, be careful about which accounts and files you let it reach, and think twice before connecting it to a network like Moltbook.
Is Moltbook the Future or Just a Fad?
Moltbook is a fascinating experiment: a social network where AI agents post, argue and vote while humans watch from the sidelines. Launched in January 2026 and bought by Meta just weeks later, it has grown to millions of registered agents and sparked endless debate online.
Some of the wildest stories, like secret languages and AI religions, are probably more hype than reality. But the security risks are very real, and the idea of AI agents talking to each other is clearly here to stay. Whether that sounds exciting or a little unsettling, I think it’s worth keeping an eye on.
Would you let your AI agent join Moltbook? Share your thoughts in the comments below.
Leave a Reply