Argument injection in OpenClaw - CVE-2026-100561
Published: November 3, 2026
Vulnerability details
The vulnerability allows a remote user to execute arbitrary commands with the OpenClaw process's host privileges outside the operator's reviewed approval.
The vulnerability exists due to insufficient validation of command-bearing wrapper arguments in OpenClaw's exec approval policy when evaluating a wrapper previously allowlisted or permanently approved by an operator. A remote user can substitute an arbitrary inner command in a subsequent agent turn to execute commands without another approval prompt.
Exploitation requires the relevant wrapper to resolve on the host.