Such a model is objectively biased regardless of where and how it's applied - it prefers X over Y, and that's bias by definition. The hard question here is deciding whether any given bias is ethical or not.