Well the AI could hallucinate the meaning of a sentence in a different way than the original, if that happens then it could be problematic.