[The Ministry of State Security Issues Warning on Multiple Security Risks of Open-Source AI Tool 'Lobster'] The open-source AI agent tool OpenClaw ('Lobster') integrates communication software with large models and autonomously executes tasks with high-level permissions. However, it poses risks such as host takeover, data leakage, manipulation of communications, and technical vulnerabilities. The article highlights that users often grant 'Lobster' the highest system permissions and entrust it with handling sensitive data. If exploited by attackers through malicious plugins or remote control, it could lead to illegal device manipulation, privacy and property damage, and even be used to spread false information or commit fraud. The Ministry of State Security recommends conducting a security assessment for 'Lobster,' adhering to the principle of least privilege, running it in an isolated environment, encrypting sensitive data, and treating it as a 'digital employee' that requires compliant, secure, and controllable management.