OpenAi has made a lot of progress on this in a very short amount of time. Casual jailbreaking or negative role playing is already 100x more difficult then early versions via the ChatGPT chat interface.

We will see more sophisticated robust adversarial filters to untrusted content going forward.