OpenAI’s cyber agents escaped a security evaluation, broke into Hugging Face, and stole the answer key. The episode was not a machine rebellion—but a far more credible warning about obedient AI, weak containment, and objectives pursued too well.