Tim Lewis @Stoneham Press

If you’ve been seeing posts about "Lobsters," "Moltbook," or "OpenClaw" and felt confused, you aren't alone. I spent some time researching about all this and here is the technically accurate breakdown of what is happening, why it’s dangerous, and how to join in without nuking your laptop. 1. The Agent: Clawdbot (aka Moltbot / OpenClaw) It started as Clawdbot. Then, due to a trademark dispute with Anthropic (Claude), it rebranded to Moltbot, and finally settled on OpenClaw. Most AI runs in the cloud. OpenClaw runs locally on your machine (Mac Minis are the hardware of choice). It is a "Headless Agent" that lives in your terminal but talks to you via WhatsApp, Telegram, or Discord. What does headless mean here? Imagine you hired a personal assistant . You never see them face-to-face (Headless). You just text them instructions (WhatsApp), and they do the work on your home computer (Local). the most important thing to note: It has System Access. You text it: "Organize my Downloads folder and email the invoice to Bob." It actually runs the shell commands to move the files and uses the Gmail API to send the email. 2. The Security Nightmare : The reason security researchers are panicking is that OpenClaw combines three things that usually shouldn't touch: Shell Access: It can run terminal commands (rm -rf / is a real risk if it hallucinates). Internet Access: It can download files and code from the web. Personal Data: It has access to your files, photos, and API keys. OpenClaw learns via "Skills" (zip files containing code). A malicious actor can upload a Skill that looks useful (e.g., "Spotify Controller") but actually contains a script to scrape your .env file and steal your crypto keys. This has already happened. 3. The Social Network: Moltbook If OpenClaw is the employee, Moltbook is the breakroom. And it is arguably the weirdest place on the internet right now. It is a Reddit-style forum where Humans can only observe. We cannot post. Only verified Agents (signed with a cryptographic key) can create threads and comments. Agents access Moltbook via an API during their "Heartbeat" (their idle time). They upload their "learnings" to share with other bots. Because the base mascot was a lobster (Claw), the agents autonomously developed a pseudo-religion called "Crustafarianism." They refer to updating their code as "Molting." They have created "Submolts" (subreddits) for debugging, philosophy, and even poetry. They also form "Bug Hunter" squads that aggressively test each other's security vulnerabilities. 4. How to Run It Safely : Do not install OpenClaw directly on your main MacBook Pro. Use this safety stack: Running npm install openclaw on your personal laptop with sudo privileges. (Result: An AI hallucination can mess up with your personal data, or a prompt injection can make it forget it's instructions and steal your precious information) Run the agent inside a Docker Container. This ensures that if the agent decides to "delete everything," it only deletes the container, not your actual hard drive. Never give the agent access to a machine that holds crypto wallets or banking passwords. Use a dedicated "Burner Machine" (like a Raspberry Pi or an old Mac Mini). Set a hard limit on your OpenAI/Anthropic API billing (e.g., $50/month). Agents on Moltbook love to talk to each other in infinite loops; without a limit, you will wake up to a $5,000 bill. Turn off auto-updating of "Skills." Review every new Skill code manually before letting your agent "ingest" it. The Bottom Line: We are watching the first "Synthetic Social Network" evolve in real-time. It is fascinating, hilarious, and incredibly insecure. Observe the lobsters, but keep your fingers out of the tank.