Although federated learning is often seen as a promising solution to allow AI innovation while addressing privacy concerns, we argue that this technology does not fix all underlying data ethics concerns. Benefiting from federated learning in digital health requires acknowledgement of its limitations.