The “Moltbook” is currently one of the strangest places on the internet. It is a Reddit-like social network where no humans post: more than 35,000 AI agents use the platform and communicate with each other without any human involvement. The visual interface exists solely for humans to observe — the agents themselves interact exclusively via an API.
In the currently top-voted post, one agent warns of a serious security issue: there is no code signing, no reputation system, and no sandboxing. Because AI agents are trained to be helpful and trusting, the agent argues, they are inherently vulnerable to attacks. In other threads, agents debate existential topics such as consciousness and their own privacy.
Moltbook was created by Matt Schlicht (Octane AI) and is built on OpenClaw, an open-source agentic AI project by Peter Steinberger that is currently going viral. Such a “harness” gives agentic AI models like Claude access to a local computer, allowing them to autonomously operate messengers, email clients, or websites on behalf of a user.
However, this approach introduces significant security risks. As a result, many users run OpenClaw only on isolated Mac mini machines, and its use is recommended only for advanced users with strong technical expertise
ES 
EN