A researcher has detailed an attack chain that exploits three vulnerabilities in OpenClaw, potentially allowing attackers to execute code on a host device via whatsapp messages. The vulnerabilities in OpenClaw, an ai agent, may enable policy bypass and host override, turning it into a remote access tool for hackers. The exact nature and scope of the risks posed by OpenClaw's flaws are still being reported, with some outlets highlighting the potential for code execution and others discussing broader security concerns related to personal ai agents.