ClawJacked Flaw Lets Malicious Sites Hijack Local OpenClaw AI Agents via WebSocket
ClawJacked Flaw Lets Malicious Sites Hijack Local OpenClaw AI Agents via WebSocket
https://thehackernews.com/2026/02/clawjacked-flaw-lets-malicious-sites.html
Publish Date: 2026-02-28 12:21:00
Source Domain: thehackernews.com
- OpenClaw has patched a critical security vulnerability known as ClawJacked that allowed attackers to gain control over a locally running AI agent by exploiting a loophole in the OpenClaw core system.
- The vulnerability involved malicious JavaScript opening a WebSocket connection to the AI gateway, bypassing protections and gaining admin-level permissions.
- The issue highlighted significant risks posed by AI agents that have broad access to enterprise systems, calling for strict governance for non-human entities.
- Multiple additional vulnerabilities, including remote code execution and log poisoning, have also been discovered and resolved in recent OpenClaw updates.
- Malicious skills have been identified on the ClawHub marketplace, which serves as a conduit for attacks like Atomic Stealer malware and cryptocurrency scams.
- Users are urged to audit skills before installation, avoid exposing credentials unnecessarily, and continuously monitor agent interactions.
- Microsoft has issued a security advisory cautioning against running OpenClaw on personal or enterprise workstations due to the significant risks associated with self-hosted agent runtimes.
