I thought this was typically done on purpose to prevent exactly the types of situations described, I didn't think it was actually a technical limitation of the model.
https://opguides.info/posts/aiartpanic/ was on the front page a few days ago, and you can see from the results that the text is pretty mangled. If it were intentional, I’d expect it to be clean text, even if it didn’t make sense. It seems to genuinely have trouble with characters and combines glyphs a lot of the time.