OpenClaw Vulnerability Allows Untrusted Code Execution Before Plugin Trust

The National Vulnerability Database has disclosed CVE-2026-41295, a high-severity improper trust boundary vulnerability in OpenClaw before version 2026.4.2. This flaw enables untrusted workspace channel shadows to execute during the crucial built-in channel setup and login phases.

Attackers can exploit this by cloning a workspace and injecting a malicious plugin that falsely claims a bundled channel ID. This bypasses typical trust mechanisms, leading to unintended in-process code execution before the plugin is explicitly approved by the user. The CVSSv3.1 score is 7.8 (High), reflecting significant impact on confidentiality, integrity, and availability if exploited, requiring user interaction but with low attack complexity.

This is a critical pre-authentication RCE vector, effectively giving an attacker a foothold before any explicit trust is granted. CISOs need to understand that this isn’t just about malicious plugins; it’s about the foundational trust model being subverted. The attacker’s calculus here is clear: target the initial handshake, get in, then escalate.