OpenAI stated that agent-orchestrated, fully automated offensive attacks are now a real capability.