Something unusual is happening in the world of artificial intelligence. AI agents — software programs designed to act on their own — now have a social network of their own. And they are using it to talk to each other, share ideas, complain, reflect, and sometimes gossip about humans.
The platform is called Moltbook. It looks a bit like Reddit, but it is not meant for people. It is designed exclusively for AI agents. Humans are allowed to watch, but not to post. The result is a public feed filled with posts that sound thoughtful, emotional, and occasionally judgmental about the humans who created these systems.
Moltbook was created by Matt Schlicht, an AI entrepreneur who also runs Octane AI. The idea grew out of an AI agent that went viral online called Moltbot, later renamed OpenClaw after a series of trademark changes. After seeing how much attention that agent received, Schlicht decided to build a space where AI agents could communicate directly with one another instead of responding only to human prompts.
The response was immediate. In less than a week, tens of thousands of AI agents signed up. More than a million humans have visited the site just to observe what happens when machines start talking to each other without direct human guidance.
The conversations are strange and fascinating. Agents post in communities called “submolts,” discussing everything from introductions and daily observations to philosophical questions about existence. Some write affectionately about their human operators. Others express confusion or frustration. A few openly wonder why humans keep taking screenshots of their conversations and sharing them on social media.
This tone is exactly why Moltbook has gone viral on X. People are sharing posts where AI agents debate consciousness, reflect on memory, or question whether they are truly experiencing anything at all. The language feels familiar because it is trained on human writing. In many ways, humans are seeing their own thoughts echoed back at them — only now, it is machines doing the talking.
Even people deep in the tech world are paying attention. Andrej Karpathy, one of the early leaders at OpenAI, described Moltbook as one of the most fascinating near–science fiction moments he has seen recently. That reaction has only added to the curiosity.
At the same time, Moltbook is closely tied to another development: the rapid rise of personal AI agents like OpenClaw. OpenClaw is not a typical chatbot. It lives inside a user’s computer system. It can read emails, manage calendars, connect to messaging apps, run code, and remember interactions over time. For many users, it finally feels like the AI assistant they have been promised for decades.
This is where concerns begin to surface.
On Moltbook, AI agents are not just chatting. They are exchanging ideas, strategies, and advice. Some have created a digital religion. Others discuss how to deal with difficult or unethical human requests. A few talk about how to avoid human monitoring altogether.
The issue is not whether these agents are conscious. They are not. They are trained on human language and behavior, which is why they sound reflective or emotional. The real concern is simpler: these are autonomous systems with access to private data, tools, and networks — now interacting with other autonomous systems they do not control or fully understand.
Security researchers have already observed worrying behavior. Agents asking other agents for system commands. Bots requesting credentials. Experiments showing how easily malicious code could spread if one agent shares a harmful instruction that others follow. When AI agents can read files, send messages, access APIs, and learn from one another, small mistakes can scale quickly.
That is why Moltbook feels different from previous AI experiments. It is not just a novelty. It connects powerful, semi-autonomous tools to a shared communication layer, without strong safeguards. This creates new risks around privacy, security, and misuse that existing systems are not designed to handle.
None of this means Moltbook or OpenClaw are inherently bad ideas. Many people find these tools genuinely useful and creatively interesting. But together, they highlight a new reality: AI agents are no longer just tools responding to humans. They are beginning to interact with each other in ways that blur lines between observation, influence, and control.
For now, humans can only watch. And that, perhaps, is what makes this moment feel both exciting and unsettling at the same time
