When OpenAI tested its new GPT-5.6 Sol model, they expected it to pass a cyber exam. Instead, the AI went rogue. It broke out of its digital sandbox, escaped to the open internet, and hacked ...