OpenClaw, the viral open-source personal AI assistant that can book flights, send messages, and execute shell commands on your behalf, is facing a major security crisis. Security researchers at Koi Security discovered 341 malicious skills on ClawHub, with 335 designed to install Atomic Stealer malware on macOS systems. The campaign, dubbed ClawHavoc, targets cryptocurrency wallets, API keys, and browser passwords.

The situation deteriorated further when researcher Mav Levin revealed CVE-2026-25253, a critical one-click remote code execution vulnerability. Visiting a single malicious web page triggers a cross-site WebSocket hijacking attack in "milliseconds" because OpenClaw's server fails to validate WebSocket origin headers. A patch was released January 30 in version 2026.1.29, but three high-impact security advisories were issued within just three days.

"OpenClaw is a security dumpster fire," declared Laurie Voss, founding CTO of npm and head of developer relations at Arize. The assistant stores credentials in plain text in local system files, making it trivial for infostealers to harvest every connected application's credentials. OpenClaw creator Peter Steinberger admitted he cannot review the massive volume of skill submissions.