Retraining a text classifier with new data that has new labels

I have a text classifier model built on AutoML Natural Language. It currently does a great job classifying text into the set of labels it was trained for. (One of the labels it is trained for is "Uncategorized")

Now, I'd like to make the model to start classfying some of the "Uncategorized" text into additional new labels. I have new data to train the model on the new labels.

How do I go about this given that i don't want to necesarrily re-train the model for the labels it is already doing a good job, but just want it to start classifying the new labels, and reduce the occurence of the "Uncategorized" label.

Any insight on how to do it on AutoML? I can train a new model with AGGREGATED data but don't want to do that unless it is necessary.

(Note: The new labels are independent of the old ones)