The OpenAI team discusses the challenges of preventing GPT from being hacked or jailbroken and the importance of giving users control within broad bounds to minimize the need for jailbreaking.