Skip to content

One-Click Agent Takeovers (ClawJacked)

Remote Code Execution via Malicious Website

Researchers discovered that malicious websites could execute arbitrary code on systems running OpenClaw through specially crafted webpages. This vulnerability allowed attackers to take complete control of AI agents with a single click.

ASF Protection

  • Sandboxed execution with minimal privileges
  • Fine-grained tool access controls
  • Network segmentation prevents lateral movement
  • Activity auditing on all tool invocations

Learn more about ASF