While I fully agree, we shouldn't anthropomorphize the models, it's also silly to pretend that "develop biases" is understood as implying anthropomorphic features of the thing being discussed. Organizations and abstract bodies develop biases, even datasets are often said to have "developed biases".