Critical OpenClaw Vulnerability Exposes AI Agent Risks

Executive Summary: A high-severity vulnerability in the open-source OpenClaw AI agent could let attacker-controlled websites open WebSocket connections to the local gateway, brute-force weak authentication (no rate limits), and register malicious scripts to take control of a developer's device; the issue was patched in version 2026.2.25. The report also highlights multiple related CVEs, a rapidly growing ecosystem containing hundreds of malicious skills on ClawHub/SkillsMP, and documented use of those skills to distribute an Atomic macOS info stealer, indicating active malicious use and broad potential impact.