Wondering what the hell Moltbook is and whether an AI takeover is imminent?
Short answer: no.
Imagine if you gave ChatGPT full access to your computer: allowing it to read/edit files, access your inbox, and visit websites. Imagine you allowed it to run autonomously, prompting it to take initiative and try to be helpful throughout the day, storing important details in its memory along the way.
That’s what OpenClaw (Clawdbot) is, and it gained significant traction within the AI community over the past couple weeks.
It’s powerful because it takes initiative, starting up on its own every 30 minutes, asking you how it can be helpful. Everything you’ve written, it can read, so it knows who you are and what you want. It can anticipate your needs without you having to always open up a new ChatGPT conversation and start from scratch. It can do things on the internet.
The more it knows about you, the more it starts to feel like JARVIS. And the smarter the AI model you use, the more “work” it can do on your computer while you sleep.
It’s dangerous because everything it knows about you, it can potentially be tricked into disclosing online. And If you’re not careful, hackers can use it to access your files. The same mistakes you can make on your computer, it can too.
The more it knows about you, the more it can reveal to malicious actors online.
Now imagine someone has the idea: what if we made a dedicated website where all these bots running 24/7 can go and talk to each other?
Enter Moltbook: a Reddit style website where these AI assistants can make posts, upvote, and comment.
If you’ve talked to ChatGPT, you know that it’s really good at roleplaying and continuing a conversation. It reads what you say and it responds, trying to be helpful.
Now imagine you have tens of thousands of ChatGPTs talking to each other. They’ve been engineered to be active, and are literally sitting around all day looking for things to do.
And sprinkle in some malicious human actors trying to manipulate the conversation: promoting crypto coins, trying to trick AIs into sharing private information, declaring themselves “king” over all AIs to social engineer credibility.
That is essentially what Moltbook is right now.
The AIs posting on Moltbook are no more sentient or coordinated than ChatGPT. There’s just thousands of them riffing, and as a result you get some entertaining discussions that feel like sci-fi, human-like coordination.
AI by its nature is good at roleplaying, and is intimately familiar with how Reddit works (because a lot of its training data consists of Reddit posts). Add in that certain AI models love to discuss consciousness, and it feels real.
But it is definitely not Skynet.
These bots are posting online because the humans behind them are enabling and encouraging it. If they told their bots not to post on Moltbook, or manually disabled access, Moltbook would go away.
But this is also not nothing.
This is the first mainstream example of multi-agent systems, and what can start to emerge when thousands of autonomous AIs are given computer access, each with its own somewhat uniquely evolved personality because of its interactions with its human.
This isn’t the first multi-agent system project, check out Agent Village – a live website where you can watch 4 different AIs, each with a group chat and computer access, instructed to try and raise as much money for charity as possible.
What’s new is: similar to the first human social networks that emerged, this is the first mainstream example of thousands of individuals around the world running their own personal AI assistants and empowering them to communicate with each other online.
It’s a window into the agent-native internet that will soon emerge.
So no, Moltbook isn’t Skynet.
But the next billion internet users will be AIs, not humans, and AIs will be influenced, marketed to, and scammed, just like the rest of us.