A few years ago there was a US Congressional Hearing [1] about how to handle deepfakes (OpenAI had a representative on the panel).

There was some deep discussion about that topic.

For example: even if all authentic content had an embedded steganographic watermark, how do you reliably authenticate recordings of recordings or otherwise degraded copies of authentic content?

[1] https://intelligence.house.gov/news/documentsingle.aspx?Docu...