Fallback result if classification not clear

Im using a custom fine tuned ada model for classification,
I was wondering is there a way to add a fallback classification result if none of the classification labels return a clear result ?
For example I use a prompt which differs from the training data and set labels which
returns a wrong classification for that prompt
in this case it should return a classification label like 4 == unknown

Did you provide examples for that in the fine tune data?

Im not quit sure how the examples should look like. I thought about picking random sentences which are not used / labeled yet. I dont know what gpt3 is gonna do with it because they have no underlying pattern
I guess I just need to test it.

Turn the question around. What would you like GPT3 to output in those cases? Pick that and write a couple of examples. You are essentially teaching it how to deal with similar situations.
Maybe the output could be just “undefined” (which is really nothing more than just another label)