The China National Security Bureau has released a "Lobster" Secure Farming Guide

On March 17, China’s Ministry of State Security (MSS) issued guidance on **OpenClaw** — the recently popular open-source AI assistant nicknamed “Lobster” — warning that while high-privilege AI agents like this can boost efficiency, they may also carry serious security risks (including device takeover, data leaks, and information misuse). Users are advised to use it responsibly and step up security protections. ### Key Details from the Guidance: The AI tool integrates communication software with a large language model (LLM) and has **high-level system access**, allowing it to directly execute user commands (shifting from “providing suggestions” to “automatic task execution”). Its plugin system supports file management, email drafting, scheduling, web browsing, and it includes long-term memory + proactive task-triggering capabilities — leading some users to call it “raising a lobster.” ### Noted Security Risks (Improper Configuration): - Remote device takeover via high-level operations - Theft of sensitive data - Exploitation of social media accounts to spread false information - System attacks through malicious plugins ### Recommended Mitigations: 1. Follow the **principle of least privilege** when running the AI agent 2. Encrypt sensitive data 3. Deploy in isolated environments (e.g., virtual machines, sandboxes) 4. Regularly audit plugin sources, access permissions, and system logs 5. Avoid exposing core devices directly to public networks All content is in English, aligned with U.S. language habits (concise, action-focused, standard tech terminology).