>>939334053 (OP)
Chat models are based off datasets and training. Even if someone actively didn't try to add their biases in, they're gonna choose data they like or deem "worthwhile"
Then they have to train it, which would lead to more sunconscious bias being added in.