This is to be expected. AI will model itself after the data that was entered in. And since that data is sexist, racist, misogynistic, and otherwise, the outputs will skew that way too.
This is the most obviously damning reason why AI cannot possibly be useful as it is today.
machine learning also in practicality helped to solved protein folding sequences. every technology in history has good and terrible uses. It's like how movable print contributed to great things and also helped the Protocols of Zion and Der Sturmer come into existence. As always it is a question of ownership and socio-economic structural incentives
Similarly, the cotton gin could have been strictly a societal improvement but the slavery industry made slavery worse by expanding with the cotton gin's use, and that persisted until slavery was (mostly) abolished.