Don't Let the Claw Grip Your Hand: A Security Analysis and Defense Framework for OpenClaw

Code agents powered by large language models can execute shell commands on behalf of users, introducing severe security vulnerabilities. This paper presents a two-phase security analysis of the OpenClaw platform. As an open-source AI agent framework that operates locally, OpenClaw can be integrat...