I haven't used it much, but I've seen some coverage of how ChatGPT and other similar tools have clearly been trained to be biased in certain specific ways. This is often around the sorts of protected group identities that American politics is obsessed with.