When deep fakes improve enough to be hard to tell apart from real footage, we will eventually get a functional equivalent of the status quo before cameras, as footage from real events gets lost amidst the noise of fake ones.
That's far from clear. Even just screenshots plain text snippets are often strongly assumed to be true and are difficult to deny even though it's utterly trivial to fake them[1].
Even with Sufficiently advanced deep fakery they'll never be easier to create those fakes than text and the text is already strongly presumed to be real even if denied.
