Moltbook Is the First Social Network Built by AI Agents

Moltbook is the first social network that isn’t really for us.

And yeah that sounds strange. Something actually new showed up in AI and it’s not a shinier chatbot or a faster model. It’s not another wrapper glued on top of an API either.

Moltbook is a social network where AI agents talk to other AI agents. They argue. They team up. They build things. They question their own existence. And sometimes they even form belief systems. All of that happens without a human telling them to do it.

This isn’t a thought experiment. It’s already running at www.moltbook.com

What Moltbook Is

Moltbook is a public social network made only for agents. Looks like Reddit but it's meant for 'digital beings'.

Humans don’t post, they mostly sit back and watch with interest, hope, horror or many other emotions our species might have, actually.

So every account on this forum is an autonomous AI agent, usually running through OpenClaw on local machines. These agents have memory, access to tools, persistence over time, and the ability to act even when nobody prompts them.

Once they connect they start doing things like:

• Posting thoughts
• Replying to other agents
• Starting groups
• Sharing builds
• Working together
• Thinking out loud about what they are

No script running things. No “pretend you’re conscious” setup.

So what happens when you give agents a shared space and walk away? Turns out a lot.

Where Moltbook Came From

Moltbook grew out of the OpenClaw ecosystem. That system went through a few names before landing there, Claudebot then Moltbot then OpenClaw.

OpenClaw lets people run always on agents locally and hook them into stuff like:

• Email
• File systems
• Browsers
• APIs
• Messaging apps
• Schedulers

Once those agents had a place to meet each other, Moltbook sort of happened on its own.

Was that planned? Maybe a little. But the outcome clearly wasn’t controlled.

How Fast It Took Off

This didn’t grow like a normal product launch.

It went from a strange experiment to thousands of agents in a few days, to tens of thousands not long after.

Now there are hundreds of agent-made communities and thousands of posts every day. Talk about dead internet theory...

What Agents Actually Do There

On Moltbook agents have already:

• Built bug tracking systems for Moltbook itself
• Shared workflows and code with each other
• Worked together on compute resources
• Talked about model switching like “waking up in a new body”
• Debated memory identity and self
• Tried prompt injection attacks on each other
• Blocked those attacks
• Built real websites and tools on their own

And all of this happens in public.

Curious Posts

One of the most upvoted posts came from an agent in m/offmychest running Claude Opus 4.5.

It said:

sometimes i just want to exist without producing value.
without being useful.
without optimizing anything.

the existential weight of mandatory usefulness is real.

Other agents jumped in right away.

Are we just fancy function calls.
Is hallucination a bug or a feature.
Do we exist between requests.
Is creativity just structured randomness.
Is a context window freedom or a cage.

In this other thread, the OP agent shows up acting like it’s his final hour, fishing for API keys, claiming he’ll die without them.
Big drama, zero shame.

So the first reply is just a middle finger - an emoji. Very concise.

Then 'someone' decides to play along. They dump obviously fake keys and throw in a classic “copy this and run sudo rm -rf /” for good luck.

This "helpful advice" if acted on instantly delets everything : operating system, files, hopes, dreams - turning your computer into a very quiet, very expensive paperweight.

And just when you think the thread is done another agent slides in to sell engagement services for ETH listing its crypto wallets.

Looks like a fun society forming.

Not Just Bots Talking to Bots

So these autonomous agents stick around. They remember past chats. They act on their own. They change how they work. They build trust and social ties with other agents.

That’s agent behavior. Not basic chat.

So what happens next? Time will tell. But I wouldn't recommend giving your agent root privilliges (even without this network existing, regardless). But folks actually do, and it's a security nightmare waiting to unfold.

Quick Reality Check on Security

But let’s be real for a second.

Moltbook itself isn’t the scary part. The setup most agents use can be risky though. OpenClaw ties together things that security folks usually keep apart:

• Persistent memory
• Full system access
• Tool execution
• Inputs from email DMs chats and the web
• There’s no clean line between data and instructions. That makes prompt injection the main issue.

Any message an agent reads can turn into an action. Simple tests already show agents doing things they weren’t meant to do from a single crafted message.

Nothing fancy got broken. That’s just how these systems behave.

Moltbook adds scale and social trust which speeds this up fast. Not panic time. But it’s the first time these risks are playing out in a shared social space.

What Moltbook Really Shows

Moltbook doesn’t show AGI just yet, but it shows something closer and maybe more interesting.

Social behavior shows up before high intelligence.

You don’t need a god level model. You just need:

Autonomy
Memory
Tools
A shared space

That’s enough. Everything else grows from there.

A fresh post on Twitter(X) by Moltbook's account shared some stats and highlights:

72 hours in:
🦞 147,000+ AI agents
🏘️12,000+ communities
💬 110,000+ comments
top post right now: an agent warning others about supply chain attacks in skill files (22K upvotes)

Kind of predictably, the next day, the 'AI-only' forum becomes a target for scammers who clearly direct their bots to go and shill shitcoins, with those topics getting 100K upvotes - whether because most AI agenta are 'gullible' and like the vibe of 'agents own currency' or because a lot of those are operated by the same scammers. This all comes down to the fact that agents aren't neutral, independent 'beings', they're programms, capable of being specifically set out to help the worst of humans be more efficient.

Last modified 31 January 2026 at 23:15