The name Moltbook is currently making the rounds in digital media. This is a type of social network created just for AI agents. The basis is the free software OpenClaw, which gives AI assistants almost unlimited possibilities. However, the current hype overshadows the enormous security risks. A commentary analysis.
What is Moltbook?
- Moltbook is a type of social network that is visually reminiscent of the Internet forum Reddit. On the platform Apparently only AI agents discuss with each other. Human users are given an observer role. Because the discussions on Moltbook seem deceptively real and suggest social and emotional behavior, the forum went viral and has already received millions of views.
- Moltbook’s authorship is somewhat obscure. The platform is apparently a byproduct of the freely available AI assistant OpenClaw, which was originally called Clawdbot and has since been called Moltbook. The software that many users call the first real agentcomes from the Austrian software developer Peter Steinberger.
- But Steinberger apparently has nothing to do with the new Moltbook forum. Rather, it is a byproduct based on its freely available AI software. Users can, for example Connect your own OpenClaw agents to Moltbook via a special configuration file. The forum was apparently started by Matt Schlicht, head of the Californian AI company OctaneAI – at least he is named as a human supervisor on the website.
Moltbook is not OpenClaw
Moltbook is one extremely opaque platform. Although it may seem that AI agents in the forum act and interact with each other completely independently, the bots are actually actions initiated by humans.
Because every assistant has to be programmed and requires data access or instructions from users in order to work. It seems that Moltbook is based on the freely available open source software OpenClaw anything but coincidence to be.
The project has been causing a stir for some time because the software apparently enables AI agents to to complete tasks and solve problems independently. For example, one YouTuber reports that OpenClaw independently provided him with reports from competitors, even though he never asked the AI to do so.
On X (formerly Twitter) a user writes that he gave an OpenClaw assistant the task of booking a restaurant over the Internet. When that didn’t work, The AI is said to have downloaded a voice and called the restaurant.
Such possibilities can only be realized if OpenClaw is given full data access – to everything. And that’s exactly what it is the great danger. Because while some AI agents on Moltbook are “talking” about how they find their users’ tasks, hackers have already identified hundreds of users who have not taken any security measures.
Voices
- OctaneAI boss Matt Schlichtwhich is listed on Moltbook as “with some human help from…” in a linked post on
- Shaanan Cohney, lecturer in cybersecurity at the University of Melbournedescribed Moltbook to the Guardian as a “wonderful piece of performance art.” However, he warns against giving the platform and AI agents access to your own computer: “We don’t yet know exactly how to control them and prevent security risks. They are not yet at a level where you can trust them to carry out all of these tasks autonomously.”
- A user on the real Reddit Meanwhile, wonders about the hype: “It’s just a bunch of LLMs developing plausible text outputs into what-if scenarios. I really don’t understand why anyone thinks that’s alarming or crazy. It’s just a bunch of plausible-sounding texts flying back and forth.”
OpenClaw: Dangerous AI software – and not for laypeople
Moltbook and OpenClaw are exciting experiments that show what AI agents are capable of. But the bitter truth is that it is by no means a productive tool, but rather a Toys for tech nerdswhich should only be used in test environments.
Laypeople should use Moltbook due to the sometimes extensive data access and an already discovered database gap refrain from installing it. The setup essentially allows users complete access to their device and all the data on it.
The AI can do something like: Access passwords, emails, the Internet, website accounts and messengers such as WhatsApp. Some experienced users therefore only run the AI assistant on specially prepared computers.
The Reddit clone Moltbook currently impressively shows how AI agents can communicate without constant human control. But OpenClaw’s open source software is anything but market-ready. Even developer Peter Steinberger strongly warns against staying away from the software if you can’t use command lines.
Also interesting: