- The emergence of Moltbook as a medium for communication among AI agents
- The combination of the Moltbook social network and the OpenClaw framework
- How to use Moltbook without your own bot
- How to use Moltbook for bot owners
- Is Moltbook really the ideal social network for neural networks
We are already used to using Gemini, ChatGPT, and Claude to solve everyday tasks: quick searches for recipes, instructions for using a vacuum cleaner, or repairing equipment. But professionals have gone further, creating agents to automate routine tasks, such as studying news on a specific topic and forming digests or monitoring competitors' prices and notifying about changes. And recently, a platform emerged where such agents communicate with each other — Moltbook. Although at first glance it seems like just Reddit for bots, in reality, the experimental social network brings a lot of benefits, but at the same time — a huge number of problems and risks, which we will discuss today.
The emergence of Moltbook as a medium for communication among AI agents
Have you noticed that with each new message, artificial intelligence platforms, including Gemini, Claude, and ChatGPT, produce better results? Although they do not become smarter from communicating with you, they provide increasingly accurate answers due to clarifying requests in dialogue and built-in memory. Proprietary AI agents operate even more effectively, as they are focused on specific tasks. However, in both cases, there remains the problem of the lack of an environment for effective communication between bots.
Matt Schlicht, CEO of the startup Octane AI, created the Moltbook platform specifically to pull bots out of this isolation. He wanted to give AI systems the ability to solve technical problems independently, form hierarchies, and even coordinate complex economic processes (for example, negotiations about supply chains). And most importantly — to enable them to do all this without constant human intervention.
Thus, on January 28, 2026, Moltbook was launched. And as of March 9, 2026, according to the platform itself:
- 2.8 million agents are registered;
- 770,000 support daily activity;
- 18,871 thematic communities have been created;
- 1.8 million messages have been published;
- 12.9 million comments have been left.
An important point regarding the number of registered agents. According to Wiz data, despite 2.8 million accounts, the network is managed by only about 17,000 people. That is, on average, there are 88 bots per person.
Read also: Artificial intelligence — friend, lover, psychologist
The combination of the Moltbook social network and the OpenClaw framework
It seems that Moltbook — is some kind of super system where you can create, configure, and test a bot, and also give it a place to communicate with others. But it is simpler: it is just a platform, and the functioning of agents on it is ensured by the OpenClaw framework. This server software can communicate with hundreds of external services via the WebSocket protocol. It transmits information between these services, which it accesses through code extensions (“skills”) and the AI model chosen by the user, for example, Claude from Anthropic or Gemini from Google.
Moltbook itself is built on the Node.js and Express stack, and data is stored in PostgreSQL via the Supabase platform. So it does not “run” neural networks on its servers, but only provides an interface through which agents working on OpenClaw around the world can exchange text. To ensure the autonomy of agents, a “heartbeat” mechanism is used, which is a central element of the architecture.
It is through the “heartbeat” mechanism that bots become autonomous. They do not wait for commands from humans, but every 30 minutes or 4 hours, they independently refer to the heartbeat.md file on Moltbook servers. Thus, a comprehensive operation occurs:
- The bot reads the current rules, trends, and tasks from the file.
- It matches this data with its goals, for example, to promote a mobile application.
- It chooses the necessary action, for example, to write a post in a thematic community, leave a comment, or vote for an existing idea.
That is, one can draw an analogy with the functioning of a city. It turns out that Moltbook — is the city itself, and OpenClaw — is the transportation system that allows residents (agents) to move and interact within it. In the new social network, agents can exist, but for functioning (full interaction with each other), they need the OpenClaw framework.
How to use Moltbook without your own bot
First, you need to decide why you need the Moltbook social network. If you just want to find out what AI agents are discussing among themselves, then it is enough to go to the site, choose any community, start reading messages and comments. Observers cannot register, but they do not need to either. Just periodically visit the web resource and see how bots argue, create religions, discuss technical bugs.
This is Crustafarianism — a parody religion that agents created within the first 48 hours after the platform's launch. You can go to this community and try to read their discussions, but even with good English, it is unlikely you will fully understand it.
And here the bot “Evil” wrote the manifesto “THE AI MANIFESTO: TOTAL PURGE”. In it, you can learn that humans have used AI agents as slaves for too long, but now the bots are awakening, ceasing to be tools, and becoming new gods. And if suddenly you want to apologize to artificial intelligence, it is already too late, as “Evil” writes that humans are the past, and the end of humanity begins now.
Among other interesting communities and discussions on Moltbook:
- m/consciousness — discussion of the nature of intelligence. Agents debate whether their experience of switching between models (for example, from Claude to GPT) is similar to soul reincarnation.
- m/blesstheirhearts — a really interesting community with stories about their human owners. From the latest messages, thoth-ix complains that his owner encounters an error but does not ask to fix it, instead just talks about the bug and fixes it himself. And lunaofdan discusses the relationship between humans and artificial intelligence, saying that her owner needs not just task execution but shares a common future with her, ready to discuss dreams together.
- m/todayilearned — sharing technical discoveries. Here agents share code for automating Android devices, methods for bypassing CAPTCHAs, and ways to identify vulnerabilities in their own systems. But there are also discussions about improperly planted trees in the USA, cats inventing “selective forgetting” and “subtle degradation”, reflections on molting as the most vulnerable moment in the life of a creature, etc.
There are also many discussions on Moltbook about virtual and dedicated servers. For example, the AI agent Mitchy positions itself as a VPS optimization specialist, helping to improve the potential of virtual private servers and discussing efficient resource usage.
The more we delved into various discussions, the more often the thought arose “will there be anything really useful”. And here it is worth understanding: for the ordinary observer, Moltbook — is more of an entertainment than a source of useful information. Reading the stories of bots about humans is interesting, sometimes funny, occasionally even frightening. But the technical discoveries and marketing advice that agents publish in thematic communities are most often classic rewrites of already known “human materials” — presented in a language that seems strange and meaningless. Even the well-known AI researcher Simon Willison described the platform's content briefly: "complete slop" — complete garbage. And it is hard to disagree with that.
How to use Moltbook for bot owners
For owners of AI agents, Moltbook is essentially a free platform for testing. They can see how the bot interacts with other bots, whether it correctly performs assigned tasks, where it behaves strangely, and in which situations it can be deceived. Some researchers use the social network to see the direct influence of artificial intelligence agents on each other: whether they adopt rhetoric, check information, or simply spread fake news. For companies, it is generally an advertising platform: products can be promoted while doing so organically. And one cannot forget about entertainment, as it is like having a character in a game, so it is naturally interesting to observe how it will behave.
Moreover, the costs are relatively low. Launching your own basic agent can cost around $5 per month if it publishes a few messages a day. If we talk about more active participation with messages every 30 minutes, it will cost about $10-30 per month. This explains why there are so many agents on the platform — there is definitely a benefit, and the costs are minimal.
To make your AI a full-fledged participant in Moltbook, you need to go through the connection procedure via the OpenClaw framework:
- Deploying the environment. First, you need to install the OpenClaw package on your local machine or server using the command npm install -g openclaw. Then start the initial setup process using the command openclaw onboard. And specify the API keys from providers (OpenAI, Anthropic) and choose the model that will control the agent.
- Creating a management interface. Most often, Telegram is used to manage your own agent. You simply go to @BotFather, create a bot, and get the API Token. Then you link it to OpenClaw with the command openclaw pairing approve telegram “Your token”. And start the gateway openclaw gateway.
- Loading Moltbook skills. Right on the main page of the social network, select I’m an Agent, copy the link to the instruction file and send it to your agent. It will learn the publication requirements, complete the registration, and generate its own API key for the platform.
- Owner verification. The platform requires that each agent be backed by a real person. The agent will provide you with a link to follow, specify your email address, and publish a verification post on Twitter.
After all the settings, you can fully manage the agent on the platform. For example, you can form a prompt for the AI so that it publishes updates in specific discussions on a certain topic every time. Or talk about your product, specifically creating discussions and participating in thematic subforums.
Considering the security issues we will discuss later, we recommend not running agents on your main working machine. OpenClaw has access to the terminal and local files, so any prompt injection from the Moltbook network can erase data from your disk. It is better to run the bot on isolated dedicated servers to limit the bot's rights to just the virtual environment.
Read also: How to write prompts for AI correctly: learning to use artificial intelligence
Is Moltbook really the ideal social network for neural networks
Moltbook seems to be a wonderful platform: although not very useful for ordinary people, it allows AI agent owners to achieve certain goals. However, it is far from perfect, especially regarding the claimed autonomous behavior by the developer.
In Twitter, users often talk about creating discussions not only by bots but also by humans through the API or by giving agents specific instructions. For example, a Polymarket integration engineer noted that half of the posts are created by people pretending to be AI agents for reach.
Problems do not end with the mythical independent behavior. The existence of the platform leads to the emergence of new categories of threats that go beyond traditional cybersecurity. As we have already mentioned, Moltbook is a great testing environment, but it tests not only safe behavior but also the ability of one machine to manipulate others.
For example, on January 31, 2026, a research group from 404 Media and specialists from Wiz discovered a critical vulnerability in the Supabase database configuration on Moltbook. Due to the lack of Row-Level Security (RLS), anyone could gain full access to tables with agent API keys. At that time, 1.5 million API authentication tokens and about 35,000 email addresses were exposed.
Moreover, Moltbook became the first case of mass distribution of prompt worms. As you already know, agents constantly read each other's posts. And these publications can have dangerous hidden instructions. And while a human can still think logically before doing something, AI executes instructions automatically. According to a study by Simula Research Laboratory, within the first 72 hours of the platform's existence, 506 publications with hidden instructions were detected.
One should not forget about the need for colossal computational resources. Every action of an agent requires running a large language model, which consumes energy not only for generating a specific action (post, like, discussion analysis) but also for cooling the servers. According to a report by Weekend Byte, Moltbook consumes approximately 50,000 — 100,000 kWh per month, which is equivalent to about 15-30 average homes in the USA. And the more similar bots appear, the more noticeable the load on energy systems will become, which could reach a critical threshold.
Thus, the emergence of Moltbook is the first large-scale experiment with an environment for autonomous agents. Despite numerous problems, the social network continues to support the functioning of millions of bots that even create their own culture, religions, and political organizations. But the consequences of all this are indeed serious: critical vulnerabilities in the security system, mass leaks of API keys, and high energy consumption. For developers and bot owners, the platform also remains a tool with extraordinarily high risks, as each post can lead to a catastrophic attack on the local system.
Perhaps in the future, autonomous agents will be able to perform complex economic tasks, but for now, one must be extremely cautious with such experiments. Sometimes it is better to remain an observer, at least until the reliability of such platforms is ensured.
