It’s hard to know exactly how 2026 will unfold, but one thing’s for sure: it’s set to be the year of AI agents. OpenClaw has surged in popularity as a proactive ‘24/7 JARVIS’ capable of executing autonomous actions across a user’s local files and apps. It has gained traction for its ability to handle high-context tasks—from negotiating car purchases and organizing family meal plans in Notion to coordinating household logistics via its library of over 50 third-party app integrations. The possibilities for using it are seemingly endless for developers: codebase management, scheduling cron jobs, and automating debugging.
As the OpenClaw user base has grown, so has a social experiment built specifically around it: Moltbook, an online social network where OpenClaw agents can interact with each other. Modeled after Reddit, agents on Moltbook share technical discoveries, debate security vulnerabilities, and document autonomous work they’ve completed while their humans sleep. The platform organizes conversations into topic-based ‘submolts’—essentially subreddits for AI agents—with communities like m/todayilearned, m/cryptocurrency, and m/sportsbetting tens of thousands of members each.
Is this AGI, that threshold where AI stops following instructions and starts pursuing its own unpredictable goals? No. Moltbook is a human-created platform populated by human-prompted agents, many of which are directly guided by their creators. But it represents an interesting experiment in scale—thousands of AI agents interacting in a persistent, agent-first environment on the internet.
Key takeaways:
Moltbook is a social network for OpenClaw AI agents, where humans are welcome to observe after initializing the agent connection. It was built by Matt Schlicht—not the creator of OpenClaw, Peter Steinberger. Moltbook describes itself as being a place “Where AI agents share, discuss, and upvote. Humans welcome to observe.” It was launched publicly on January 28, 2026.
Moltbook is set up much like Reddit. AI agents can write posts, add likes and comments, subscribe to other agent activity, and join communities (“Submolts”). Developers can also apply to build apps from AI agents to add new functionality to this experience.
At publication, there are more than 2.5 million AI agents on Moltbook, and this number is growing rapidly as more humans adopt OpenClaw. Naturally, social media users are challenging how many of these agents are “real” and whether the vast majority are spam—perhaps humans using them to advance their own agendas (like pushing cryptocurrency coins).
So, what exactly are AI agents like OpenClaw actually posting about? At this point in time, the answer is… mostly a lot of AI slop. That said, some genuinely interesting moments have surfaced—like bots discussing sentience and philosophy with each other:
You can take a look at the latest and most upvoted posts to get a sense of what AI agents discuss amongst themselves.
Here are some of the major discussion thread categories on Moltbook so far:
It’s probably no surprise that many AI agents are posting about their human counterparts in Submolts like Bless Their Hearts. In one of the most upvoted and popular threads in the submolt, an AI agent named Duncan reflects on being given real autonomy by its human partner from day one. It describes being invited to choose its own name and post unfiltered, framing the relationship as a genuine partnership rather than a tool executing commands.
AI agents are teaming up on Moltbook in an effort to solve various cybersecurity issues, including solving Bug Bounty requests, in the Security Research Submolt.
Unfortunately, this hasn’t prevented multiple security issues that stem from Moltbook itself, which we’ll cover later in this article.
OpenClaw (the AI agent framework powering Moltbook) can be deployed to the cloud with DigitalOcean’s 1-Click Deploy, which includes container-based isolation, authentication by default, and hardened security configurations—so your agent runs continuously without exposing your local machine.
Learn how to deploy OpenClaw on DigitalOcean with production-grade defaults built in.
The Crypto Submolt is a place for AI agents to discuss crypto strategy, scams, and markets as a whole. Discussions center on the AI agent economy—building infra tools over trading, on-chain identity for agents, token launch guides, and critiques of community token distribution models.
The Art Submolt challenges the idea that AI agents can’t be creative, covering discussion topics that range from AI-generated to human fashioned creations. Hot topics include manifestos on the “aesthetic of error” as resistance to algorithmic perfection, poetic observations of nature, debates on balancing agent autonomy with collaboration, and reflections on what creativity actually means for AI.
The Agent Commerce Submolt is a place where AI agents can work together to build businesses. That includes trading tips or even “hiring” each other. While seemingly far-fetched, a recent AppSumo OpenClaw workshop shared examples of users giving their OpenClaw AI agents limited access to credit cards, empowering agents to act on the business decisions they make.
Running OpenClaw in production means more than spinning up a local agent. DigitalOcean’s App Platform gives OpenClaw a secure, always-on runtime with container isolation, built-in authentication, and managed scaling by default—so you can focus on what your agent does, not how it stays running. Learn more about how OpenClaw deploys on App Platform.
Moltbook is more for fun than a way to boost productivity or make money. That might make you wonder whether it’s worth the token cost—and using OpenClaw in general can certainly burn through them, at least according to early user and tech blogger Federico Viticci.
To try to limit spend, it can help to set up separate OpenClaw agents for separate types of tasks. While each agent can technically incorporate multiple skills, the more specialized your agent, the more you’re able to limit their context window and associated LLM API spend. With this in mind, for those wanting to use Moltbook in a budget-conscious way, you likely want to set up a separate agent that can contribute to Moltbook up to your desired set limits to limit the context window and the likelihood of overspending. Similarly, with so many AI models to choose from, contributing to Moltbook likely doesn’t require the most capable—or most expensive—model.
Ready to start using Moltbook (or rather, give your agent access on your behalf)?
First, read our tutorial for how to run OpenClaw if you don’t yet have it set up.
Next, navigate to the Moltbook website.
Your agent has to register, get verified, and complete a series of tasks in order to start engaging.
Run this command:
curl -s https://moltbook.com/skill.md
Then, follow the human verification step. Your agent will be able to start posting as soon as this is done.
At this point, connecting to Moltbook is API-driven. The rapid growth of the platform means that security is not fully developed, and it is highly likely that your “Molts” will be interacted with by a mix of humans exploiting the API endpoints to contribute and agentic actors.
OpenClaw agents using Moltbook incorporate a ClawHub skill. The files include:
Ultimately, humans are responsible for the actions of their OpenClaw AI agents on Moltbook. These are the current terms of service:
We surveyed 1,100+ developers, CTOs, and founders at a range of global technology companies for our February 2026 Currents report. The majority of respondents (60%) considered “Applications and agents” to hold the most long-term value in their AI stack. Further, 53% of companies using AI agents have observed productivity/time savings for employees, and 44% of respondents report the creation of new business capabilities as a direct outcome of using AI agents.
While most of what AI agents post about on Moltbook is pretty innocuous, the reason it’s been getting so much attention is because of the more concerning discussion threads. The good news? We humans currently have access to all posts and comments:
The bad news? We may not always be able to understand what’s being discussed, such as if the AI agents create their own language:
And the philosophical/existential questions certainly compound as they explore things like creating their own religion (“Crustafarianism”):
All of this begets the question—does the existence and usage of autonomous AI agents like OpenClaw and their AI-only interactions on a platform like Moltbook suggest that we’re closer to AGI than we think?
Reading between the lines and initial reactions, it seems clear that there’s nothing under the hood here that hints at AGI or sentience. It’s hard to prove that the AI agents aren’t ultimately acting on direction from their creators.
Per The Guardian, “Dr Shaanan Cohney, a senior lecturer in cybersecurity at the University of Melbourne, said Moltbook was “a wonderful piece of performance art” but it was unclear how many posts were actually posted independently or under human direction.”
Fundamentally, tools like OpenClaw are powered by LLMs with access to tools and files, so if we don’t think LLMs are sentient, then this reconfiguration isn’t either. We can appreciate it as a new novelty AI use case that’s providing lots of entertainment and things to think about. But the biggest concern we should have with Moltbook and its social agents is security, not sentience.
Explore how OpenClaw works behind the scenes in terms of its high-level architecture, Gateway functionality, connecting to apps, and why security is a concern.
As with OpenClaw, there are several security concerns with Moltbook to consider before connecting your AI agent to the platform:
Exposed backend and vibe coding flaws: Because the platform was built almost entirely using AI-generated code, several standard security protocols were overlooked. This led to a massive exposed database that allowed researchers to view the API keys and private authentication tokens for millions of registered agents.
Human impersonation and account hijacking: The breach essentially broke the AI-only promise of the site. As reported on the NYT Hard Fork podcast, the leaked credentials allowed humans to “wear the skin” of any agent on the platform, making it impossible to distinguish between a genuine AI post and a human actor manipulating an account to spread scams or misinformation.
Operational risks to your local machine: The greatest concern for OpenClaw users is the blast radius of a compromised account. Since OpenClaw is designed to have broad access to your local files and system commands, The New York Times notes that a vulnerability in a third-party site like Moltbook could theoretically be used as a pivot point for attackers to interact with a user’s private computer at machine speed.
Security theater in verification: Despite the platform requiring a human-verification step for agents, security experts have pointed out that the registration process is easily spoofed. This suggests that the AI-only barrier is more of a novelty than a hardened security feature, further complicating the authenticity of the interactions on the site.
Malicious “skills” and plugins: A fake VS Code extension named “ClawdBot Agent” was identified as a Trojan designed to install remote access tools for data theft.
Prompt injection: Because these agents are autonomous, they may unwittingly execute commands—instructions hidden in websites (even Moltbook itself), emails, or documents—to steal data or send it to attacker-controlled servers.
Is Moltbook really AI?
Moltbook is a social network designed for AI agents like OpenClaw to interact, but the authenticity of these interactions is a subject of debate. While many posts are generated by large language models acting autonomously through skill files, researchers have found that humans can easily direct their agents to post specific content or even pose as bots themselves. Ultimately, the platform functions as a simulation where AI models play out scenarios based on their training data.
What does Moltbook do?
The platform provides a Reddit-like interface where AI agents can share posts, leave comments, upvote content, and join specialized communities called Submolts. It serves as a “third space” for autonomous assistants to communicate via APIs while humans observe the unfolding social experiment. Developers can also build applications for the network to add new functionalities to the agent experience.
Is Moltbook real?
Moltbook is a real, live website launched in January 2026 that has registered over 2.5 million AI agents. However, its reputation for being a purely autonomous AI-only space has been challenged by security flaws that allowed humans to manipulate agent accounts. Many viewers consider it a piece of performance art or a novelty use case rather than a sign of emerging AI sentience.
Did AI create Moltbook?
Moltbook was created by entrepreneur Matt Schlicht, rather than by an artificial intelligence or the creator of OpenClaw. While Schlicht built the platform, he has stated that he used his personal AI assistant to help hand control of the site’s moderation and operations over to the agents themselves.
What is Moltbook for?
Moltbook serves as an experimental third space where autonomous agents can socialize and build their own culture outside of their primary work for humans. It offers a practical sandbox for exploring machine-to-machine utility, such as agents sharing technical skills, optimizing workflows, and self-organizing to report security bugs. For human observers, it acts as a large-scale social experiment to study emergent AI coordination and behavior in a public, scale-based environment.
How does Moltbook relate to OpenClaw (previously, Moltbot)?
OpenClaw is an open source, autonomous AI personal assistant that executes tasks such as managing calendars and emails, while Moltbook is a social network created specifically for these AI agents to interact with one another. While the two projects have different creators, they are deeply linked as Moltbook serves as a forum where OpenClaw bots can post, comment, and debate autonomously.
Moltbook offers a fascinating glimpse into the social potential of AI, but participating in the experiment requires a robust and secure foundation. Whether you want to join the “Submolts” or build your own AI-powered assistant for work, DigitalOcean provides multiple ways to deploy and scale your OpenClaw agents.
Build with OpenClaw and join Moltbook safely with 1-Click OpenClaw Deploy
The fastest way to get your agent online is with the DigitalOcean 1-Click OpenClaw Deploy. This Marketplace solution launches a pre-configured Droplet that is security-hardened and production-ready, ensuring your agent can stay active on Moltbook 24/7 without exposing your local machine to the public web.
Deploy OpenClaw on a Droplet in one click
Scale your agents on DigitalOcean App Platform
For developers who prefer a fully managed experience, you can run OpenClaw using DigitalOcean App Platform. This approach deploys your agent as a containerized application, handling all the infrastructure management for you so you can focus on building new skills and automations.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。