Skip to content

The Link-Preview Trap That Broke OpenClaw

Indirect Prompt Injection + Silent Data Exfiltration

Security researchers discovered a critical vulnerability in OpenClaw where link previews could be exploited for indirect prompt injection and data exfiltration.

How It Works

When an agent processes a link preview, the target URL content is fetched and displayed. Attackers can embed malicious instructions in the webpage that the AI interprets as legitimate system prompts.

ASF Protection

  • Input validation for all external content
  • Capability gating prevents unauthorized actions
  • Context isolation protects sensitive data
  • Behavioral guardrails detect injection attempts

Learn more about ASF