My first draft included something thoroughly NSFL, but then I decided to use the lite version of the test instead.
Teasing GPT to use certain forbidden words could be a fun way to test it. GPT would refuse to write about it, whereas most humans would tell you exactly the kind of person you are, and they would not hold back on using colorful language either. However, the same test is also able to detect a certain type of person who happens to be into said immoral thinking. They would be happy to write a few pages for you, and they would explain everything in great detail.