AI Agents Show Promising Penetration Test Success Rates

A new study reveals that AI-powered agents achieved penetration-test success rates between 10.7% and 69.3%. The evaluation tested 19 models against 300 servers using general-purpose cybersecurity tools. While autonomy is advancing, these agents are not yet capable of successfully hacking real networks. Nevertheless, they can automate basic attack workflows, increasing cybersecurity risks. Organizations should strengthen defenses to mitigate this evolving threat.