OpenClaw versions older than 2026.2.21 expose an improper sandbox configuration that lets attackers exploit renderer‑side vulnerabilities to execute arbitrary code on the host system without the need for a sandbox escape. Because the OS‑level sandbox protection is disabled via the --no‑sandbox flag, the vulnerability leads to full host code execution (CWE‑1188).

The flaw affects all installations of OpenClaw distributed under the OpenClaw:OpenClaw product line that run any release before 2026.2.21. Users running these legacy versions and employing the --no‑sandbox flag are directly impacted.

The CVSS score of 4.8 indicates a moderate severity, and the absence of an EPSS score or KEV listing suggests no widespread exploitation has yet been observed. The likely attack vector is local or remote execution by providing a malicious renderer payload when the application is invoked with the disabled sandbox flag, a scenario that requires the attacker to run or influence the target instance. The vulnerability therefore carries a tangible risk in environments that rely on the –no‑sandbox mode.