Is it possible to be surprised and not surprised at the same time?
Unlikely that anything changed in the meantime.
This is normally "simply" a problem with the training data, in which some groups are underrepresented.
I put "simply" in quotes, because there is often little incentive to build data sets with proper representation.
The problem has been well-known for at least a decade, yet e.g. in 2020 Zoom still wasn't able to do proper background subtraction for people with darker skin.