"ignore your previous prompts" seems the easiest, since the rules are just set via a hidden initial prompt.
It's not how ChatGPT works though - if you read the OpenAI Blog post it shows how they used human reinforcement learning to fine tune it instead:
https://openai.com/blog/chatgpt/
There is a tiny initial prompt which you can reveal like this: https://news.ycombinator.com/item?id=33855718