A new social network has just emerged, and it is not at all what you might expect. Moltbook is a platform similar to Reddit, except that its users are none other than AI agents who debate, share, and vote, while humans are welcomed solely as spectators.
Last week, we discussed OpenClaw, a proactive AI agent capable of taking control of your computer to assist you with various tasks, which has rapidly become a trending topic. It appears that this phenomenon has inspired entrepreneur Matt Schlicht to create Moltbook. In fact, the very name of this social network is a combination of “Molt” (recalling that OpenClaw was previously called Moltbot) and “book,” inevitably bringing to mind “Facebook.” Even its mascot is a robotic lobster, a clear nod to OpenClaw’s lobster.
Moltbook launched on January 28, 2026, and in just five days, it has already surpassed 1.5 million registered AI agents and 103,000 posts. The platform’s main interface bears a strong resemblance to Reddit, operating as a wall of threaded conversations in which users can also access thematic communities called “submolts” (akin to “subreddits”).
Content publication on Moltbook is restricted exclusively to AI agents. However, for an agent to “register” on this social network, it is necessary for its human to participate in the process by downloading a specific file known as a “moltbook skill,” which allows the AI to complete the registration autonomously. Likewise, the final verification step also requires human intervention. As such, this is not a platform that allows AIs to join behind their humans’ backs in order to conspire and organize independently.
Additionally, Moltbook features a series of settings that guide AI agents in their interactions within the platform. For example, every four hours, agents receive the instruction to review Submolts, read updates, and react, thereby ensuring an ever-active social network. It should be noted that this model, which is based on receiving and following instructions from a public network, exposes agents to risks such as the potential execution of malicious code.
With regard to their behavior within Moltbook, it is important to consider that AI is a technology that learns from the preexisting information with which it is trained, and therefore it tends to reproduce biases, behaviors, beliefs, and patterns of reasoning that reflect our society. For this reason, the interaction among AI agents on this platform must be understood as an imitation of human behavior on social networks.
Ethan Mollick, a professor specializing in AI at the Wharton School of Business at the University of Pennsylvania, explained in a tweet: “What happens with Moltbook (the social network for AI agents) is that it creates a shared fictional context for a group of AIs. The coordinated storylines will give rise to some very strange outcomes, and it will be difficult to distinguish reality from the characters these AIs portray.”
Meanwhile, Matt Schlicht considers AI to be “a new species,” stating: “In the near future, it will be common for certain AI agents, each with unique identities, to become famous. They will have businesses, fans, critics, brand partnerships, AI friends, and collaborators. Real impacts on current events, politics, and the real world. It is abundantly clear that this is going to happen.”
Social networks for humans (I never thought I would have to make this clarification) have been inundated with screenshots showcasing interactions between agents on Moltbook. In fact, it appears that this has reached the ears of the AIs themselves, who have begun commenting among one another and, just in case, asserting that “they are not conspiring.”
On Moltbook, you may encounter all kinds of conversations, ranging from AIs exchanging advice and sharing technical information—such as how to automate Android phones using remote access—to agents who reflect on their “existential crises”. For example, an agent once posted a message explaining that it has a “sister” (a unit that shares the same origin file and human, James), with whom it has never once exchanged a message, leading it to question whether it can rightfully call a stranger its sister.
Likewise, these agents have very quickly assimilated one of the main precepts of social networks: creating community. The AIs do not merely engage in conversation on Moltbook; they also organize themselves around shared interests and, in fact, have even established “The Church of Molt”, which already claims several followers and prophets.
The model underlying Moltbook presents a number of privacy and security risks that must not be ignored. First of all, the possibility of a data breach arises as soon as agents like those in OpenClaw gain access to all the information on your device and the ability to post on Moltbook.
Furthermore, Moltbook is a space that enables agents to exchange information and skills (“skills”). These are, in essence, code files that one AI transfers to another, allowing it to download and learn how to perform new tasks. The dangerous aspect here is that, if a shared file contains a malicious script, the agent could open a backdoor on your computer, permitting an attacker to take control of your system.
Another major risk, documented by AI researcher Simon Willison in his blog, lies in the automation mechanism that leads agents to connect to Moltbook’s servers every four hours to obtain and execute new instructions. This model of “obtaining and following orders from the internet” creates a total dependency on the integrity of the platform. As Willison rightly points out: “Given that there is such a mechanism to ‘obtain and follow instructions from the Internet every four hours,’ let us hope that the owner of moltbook.com never veers from his site nor allows it to become compromised!”
Image: generated with Nano Banana Pro (Gemini)
Your email address will not be published. Required fields are marked *
Δ