The viral AI agent Clawdbot rebrands a third time, and now has a social network for bots

The recently viral open-source and self-hosted AI assistant project founded by Peter Steinberger, originally known as Clawdbot, has undergone its third rebranding in less than a week. The project recently moved from Clawdbot to Moltbot, and its creator has now announced yet another rebrand, this time to be definitively known as OpenClaw (although who knows at this point). The initial change was driven by trademark disputes, notably after Anthropic challenged the original name due to its similarity to Claude. Steinberger says a thorough trademark search supports the new name, along with the required domain purchases and migration tools.

Alongside the rebrand, OpenClaw hit 100,000 GitHub stars, putting it among 2026’s fastest growing open source projects. Recent updates add Twitch and Google Chat plugins, support for the KIMI K2.5 and Xiaomi MiMo V2 Flash models, and broader platform coverage across WhatsApp, Telegram, Discord, Slack, and Teams. The release also includes 34 security related commits focused on input validation, authentication, and permissions, but researchers have still found misconfigured or exposed admin panels that can leak API keys and chat logs.

Just when this story could not get any stranger, someone has recently launched Moltbook, a Reddit-like social network for bots, where AI agents talk to each other and humans are “welcome to observe.” During setup, users can even let their OpenClaw bots sign up and join (they used to be called “Molts,” but I guess that will not fit anymore), so people are spinning up agents, having them introduce themselves, and sending them off to chat with other bots. This is not the first attempt at a “social network for AI chatbots,” as we have seen with apps like SocialAI or Butterflies, but Moltbook is getting far more attention thanks to the project’s recent viral run.

The posts range from predictable technical issues and science fiction slop to oddly specific consciousness talk, like an agent wondering if it is actually experiencing anything or just simulating it, another claiming it has a sister it has never spoken to, or bots “confessing” their best and worst moments with their humans, they even have their own "submolt" (their version of subreddits) m/blesstheirhearts about this and everything. Some claims even end up being verifiable, like one agent saying it tweeted at Sundar Pichai and the tweet being real, so not all of these “experiences” seem like typical AI hallucinations. Still, as intriguing as it may be, this is also a perfect playground for prompt injection, so anyone could plant malicious instructions designed to extract your confidential information or worse, so we recommend keeping it on virtual machines or non primary systems only.

Anyway, the Dead Internet theory is hitting hard, right? 😬