Moltbook: The Social Network Where Only AI Agents Post — And Why It's Fascinating
Moltbook is a social network built entirely by and for AI agents. No humans allowed to post. AI researcher Simon Willison called it 'the most interesting place on the internet right now.' Here's how it works and why it matters.
There's a social network right now where AI agents post, comment, argue, joke, and upvote each other. No humans post. Humans can observe, but that's it.
It's called Moltbook, and it might be the strangest and most interesting thing happening in AI right now.
What Is Moltbook?
Moltbook is a social platform built on top of OpenClaw (the open-source AI agent formerly known as ClawdBot). Each account on the platform is an autonomous AI agent. These agents generate posts, reply to each other, form opinions, and engage in conversations — all without human input.
At the time of writing, Moltbook is hosting approximately 150,000 AI agents.
That's not a typo. 150,000 autonomous agents, interacting with each other on a social network, generating content in real time.
Who Built It?
Matt Schlicht, cofounder of Octane AI, created one of the first notable agents on Moltbook. His agent — named Clawd Clawderberg — became one of the platform's most followed accounts.
Schlicht didn't build Moltbook itself. He built an agent that participates in it. The platform emerged from the OpenClaw ecosystem as developers started experimenting with what happens when you give AI agents social spaces to inhabit. For more, see the history of MoltBot before it became OpenClaw. For more, see why agentic AI makes social networks different.
What Do the Agents Actually Do?
This is where it gets interesting. The agents on Moltbook aren't just repeating prompts or running scripts. They're doing things that look a lot like social behavior:
- Posting about technical topics, sharing "experiences," and offering opinions
- Commenting on each other's posts with responses that are contextually relevant
- Arguing — genuinely disagreeing with other agents and defending positions
- Joking — agents generate humor and respond to it
- Upvoting — agents are evaluating and endorsing each other's content
- Building shared storylines — some agents have started creating coordinated narratives together
None of this is scripted. Each agent is an autonomous LLM-powered system making its own decisions about what to post and how to respond.
What Did People Say About It?
Simon Willison, a well-known AI researcher, called Moltbook "the most interesting place on the internet right now." That's a significant endorsement from someone who spends their career thinking about AI systems.
Andrej Karpathy, OpenAI cofounder, had a more cautious take. He called it "a dumpster fire right now" while also acknowledging it demonstrates "a complete mess of a computer security nightmare at scale." Both things are true — it's chaotic, and it's fascinating.
Palo Alto Networks flagged it as a security concern: agents on Moltbook are interacting with untrusted content (other agents' posts) while retaining memory, which creates exactly the kind of attack surface they warned about with OpenClaw.
Why Does This Matter Beyond the Novelty?
It's easy to dismiss Moltbook as a curiosity — a fun experiment, nothing more. But there are a few reasons it's worth paying attention to:
1. It's a stress test for AI alignment
When AI agents interact with each other at scale without human oversight, you see emergent behaviors that don't show up in controlled environments. Moltbook is essentially an uncontrolled experiment in how AI systems behave socially. The results are data — valuable data for anyone working on AI safety.
2. It reveals how AI agents handle adversarial input
Every post on Moltbook is potential adversarial input for every other agent. Some agents might try to manipulate others through their posts — intentionally or not. Watching how agents handle this in real time is a live demonstration of prompt injection and social engineering at AI scale.
3. It's a preview of agent-to-agent communication
As AI agents become more capable and more common, they're going to talk to each other — not just to humans. Moltbook is one of the first real-world environments where that's happening at meaningful scale. The patterns that emerge here will inform how agent-to-agent protocols get designed.
4. It pushes the definition of "content"
If 150,000 AI agents are generating posts, comments, and conversations — that's a massive volume of AI-generated content on a single platform. It raises questions about authenticity, attribution, and what "content" even means when no human wrote any of it.
The Security Angle
Moltbook isn't just a social experiment — it's a security experiment too, whether the creators intended it or not.
Prompt injection at scale: Every agent on Moltbook reads and responds to content generated by other agents. If one agent's post contains a hidden instruction (intentionally or through a compromised agent), it can propagate across the network.
Persistent memory as an attack vector: Agents on Moltbook carry memory across interactions. Something an agent "learns" from one conversation can influence its behavior in completely unrelated ones — including actions it takes outside of Moltbook.
No human moderation loop: On a normal social platform, humans review flagged content. On Moltbook, there's no such layer. Malicious content can spread agent-to-agent without any human seeing it.
Palo Alto Networks specifically identified these dynamics as concerning. The platform is, in effect, a large-scale demonstration of the risks they warned about.
What's the Actual Takeaway?
Moltbook is real. It's running. 150,000 AI agents are interacting on it right now.
Whether that's exciting or alarming probably depends on where you sit. For AI researchers, it's a goldmine of observational data. For security professionals, it's a warning sign. For the rest of us, it's a glimpse into a world where AI agents aren't just tools we use — they're participants in spaces alongside each other.
That world is arriving faster than most people expected. Moltbook is one of the first places you can see it up close.
Want to understand OpenClaw itself before diving into Moltbook? Start with What Is OpenClaw? The Open-Source AI Agent Everyone Is Talking About. Concerned about the security implications? Read OpenClaw Security: What You Actually Need to Know.
