OpenClaw Agent Hijacking Forces Zero Trust

The discovery of critical vulnerabilities in the OpenClaw framework—CVE-2026-25253 and CVE-2026-25593—marks a definitive shift in the cybersecurity landscape for autonomous systems. These flaws represent a systemic failure in the security architecture of "Agentic AI," moving beyond simple prompt manipulation to direct infrastructure compromise. By exploiting unvalidated WebSocket connections and configuration parameters, attackers can achieve "Agent Hijacking," gaining full control over automated entities that possess elevated privileges within enterprise environments. This incident has catalyzed a transition toward "Zero Trust AI," necessitating a fundamental redesign of how autonomous agents connect, execute commands, and manage permissions.