OpenClaw hit by “ClawJacked” exploit: 40,000 AI agent systems allegedly compromised via silent web-based takeover bug, as users debate whether viral warning is real fix or social-engineering hoax

OpenClaw, a popular open‑source AI agent orchestration platform, was recently found to contain a critical vulnerability dubbed “ClawJacked” that allowed a malicious website opened in a user’s browser to silently hijack a locally running OpenClaw agent and potentially compromise the entire workstation. Security firm Oasis Security reported that any site the developer visited could open a WebSocket connection to the OpenClaw gateway on localhost, brute‑force the gateway password due to missing rate limiting, auto‑register as a trusted device without user approval, and then gain full control of the agent, including reading configuration data, logs, connected nodes, and in some attack chains executing arbitrary shell commands.

✨ AI-generated background, compiled from web sources — not editorial content.

More coverage

• ClawJacked flaw exposed OpenClaw users to data theft (securityaffairs.com)
• ClawJacked Flaw Lets Malicious Sites Hijack Local OpenClaw AI ... (thehackernews.com)
• ClawJacked: How a Single Browser Tab Achieves Full Workstation ... (securebydezign.com)
• ClawJacked Vulnerability in OpenClaw Allows Malicious Websites ... (rescana.com)
• ClawJacked: OpenClaw Vulnerability Enables Full Agent Takeover (oasis.security)

Explore the topic

• AI Agents
• OpenClaw
• AI