Models & Releases

OpenClaw security audit reveals 36% of skills have vulnerabilities, 76 are malware

Popular AI agent platform exposes 135k instances to critical security threats.

Deep Dive

Security audits reveal OpenClaw's massive vulnerabilities: 36% of 4,000 scanned skills had security flaws, 76 were confirmed malware, and 12% were malicious in community audits. Attackers deployed 335 fake skills installing Atomic macOS Stealer. Default configurations exposed 135,000 instances to the internet, while invisible email instructions could trick agents into creating backdoors. Experts now recommend Docker-only deployment with strict vetting for all community skills.

Why It Matters

Widespread adoption of vulnerable AI agents creates unprecedented attack surfaces for both individual users and organizations.

📬 Get the top 10 AI stories daily