It took a researcher fewer than 2 hours to hijack OpenClaw

A researcher found that it took fewer than 2 hours to hijack OpenClaw, a personal AI agent, highlighting significant security concerns. OpenClaw has full system access by default, can read files, execute commands, and manage credentials. Researchers have identified vulnerabilities, including remote code execution bugs and a malware-stuffed plug-in ecosystem. Moltbook, a social network for OpenClaw agents, also suffered a critical backend misconfiguration that exposed its primary database. The article warns that the lack of meaningful security in OpenClaw and Moltbook may harm unwary developers and users.