If you want to teach Face ID to reject masks, you need to make some masks. Similarly, if it needs to be taught to reject a twin, you need dozens of twins. And if it starts labelling people incorrectly as their twins, is it worth it?
Perhaps they can sidestep this by offering a specific twin learning feature.
Twins are just examples of two people with very similar faces. If Apple are able to train Face ID to distinguish between 6 billion different faces they will also be able to distinguish faces of twins.
The twin's mother is inside the phone telling it what to do? Or the person you replied to is talking about technology finding it difficult to distinguish, and not a close family member?
My point is that twins are only difficult to distinguish for people who do not know them. For family and friends it's easy, which means that there are substantial differences even for faces of twins.
The face detection technology will be able to recognize those differences as good or better than the twins' family and friends. If it does not now, it will eventually.
Perhaps they can sidestep this by offering a specific twin learning feature.