The new data types nullable integer, nullable boolean, and string cannot do boolean indexing when the columns contain missing values.
I searched the pd.NA and the boolean indexing sections and have not seen this mentioned. Users need to be aware of this limitation as it greatly reduces their usability.