A newly disclosed vulnerability in OpenClaw allowed attackers to hijack the AI assistant and execute arbitrary system commands simply by luring users to a malicious webpage. The flaw exploited weak token handling and origin checks, turning OpenClaw’s powerful local access into a full system takeover risk. While the issue is now patched, it’s a sharp reminder that agentic AI tools blur the line between software and operator—and when security lags, the blast radius grows fast.
…