The problem with AI upscaling is that it does add something. It fills in the details with things that could plausibly be there, regardless of if they are. It's especially dangerous if it's used for something like security footage, where it'll do stuff like make up a face based on a few pixels.
Depending on the context it's probably not that bad, but there's plenty of details in youtube videos that people pay attention to, like in news, history, tutorials, educational content, and so on. Even for a fictional story, it could add nonsense that people assume is part of the actual show.
If the original footage is so bad that "nonsense that people assume is part of the actual show" "could plausibly be there", then the problem is not with the AI... it wouldn't be the first time I'm confused by the artifacts in a low quality video.