This.
The bias in generative AI is not a reflection of bias in society. Why would it be? It's a reflection of bias in the training data, which of course is very very far from a complete representation of the world. The training data is whatever the makers decided to throw in there and often even they have no idea how the soup is made or how well it represents any society or culture. And generative AI will then tend to amplify and exacerbate bias, because... maths.
Bonus insight: if you try to mitigate bias with inclusive prompting, it's not always that simple.