Command injection in OpenClaw - CVE-2026-27001
Published: May 4, 2026
OpenClaw
Detailed vulnerability description
The vulnerability allows a remote user to disclose sensitive information or influence agent behavior.
The vulnerability exists due to improper neutralization of special elements in the agent system prompt when embedding the current working directory path into LLM prompts. A remote user can cause OpenClaw to run inside a specially crafted directory name to disclose sensitive information or influence agent behavior.
The crafted directory name may contain control or format characters, including newlines or Unicode bidi or zero-width markers.