At the very least, responsible businesses can agree by common convention to abide by this code of conduct. There could be something like a "good AI seal of approval" issued to businesses that voluntarily adhere to the standard, like those web certificates that were ubiquitous in the early Aughts. Not only should the communications be public-key signed so the browser can identify them, but the the synthetic authors/speakers etc. should be named in a consistent way that unequivocally identifies them as AI.
For example, appending a numeral to the name. Instead of calling itself Joe Rogan, the AI would be "Joe 7 Rogan." Or instead of an article written by Jennifer 8. Lee, it would be written by, hmm, okay there might be a little problem in her[0] case. One thing I've been mulling around is the idea of repurposing the Apple][ symbol. So Joe][ and Jennifer][, where ][ is pronounced "two." The convenience being that it's comprised of glyphs that are on every keyboard, and they also look mechanical and nothing like normal names.
How do you even set safeguards for this?
I think a good analogy is how technological advances made forging physical currency trivially easy. Governments that deployed anti-forgery measures eventually weathered the storm. Governments that didn't ended up with catastrophically weak currencies. Information and data are the ultimate currency on the Internet.