The go-to example of a failure state in the golden rule that I am aware of is "thirsty person wants everyone else to jump their bones, thus the rule suggests that they should jump everyone else's bones".

The take-home being, as long as peoples preferences are different, the rule has some failure states based upon that.

And the wider example that this points to is that as long as everyone's values are different.. the rule loses a lot of it's luster. The more different another person's values are, the less you can rely on projecting your own values as a method to gainfully interact with them.

In place of that biblical aphorism (born from a monoculture no less), there is a lot of room available for us to build something more robust and sophisticated and the AI alignment problem offers a fine excuse to get cracking and try to formalize something already! ;)