Others have found more subtle ways to fool the AI into giving problematic responses.
click here