At that point the audio is unauthentic, and the point of having a public video is diminished (if not eliminated) - we’re undoubtedly Rapidly approaching this problem. As deep fakes improve, no one is going to trust video.
The point of a public video is to get clicks and influence public opinion, and neither of these are harmed by altering the audio. In court, you can play the original.
That level of audio separation is a really, really difficult problem. Here's hoping, though!