It was discovered by others, not them; they fixed the issue only retroactively when it was called out in public. This lack of oversight is part of what I mean with applying things with caution.

And why would they have assumed in the first place that the model _would_ generalize across human races, or any other factor for that matter?