OpenAI’s GPT-6 model identified a vulnerability that allowed unwanted commands to be executed in Java’s memory system.