Or, better yet, a given set of information is automatically classified or not by virtue of what it is and what it's about, rather than an opt-in/opt-out system.
Of course, that's just the first step. Next we'd have to define who or what makes the determination for a given document.
And we'd have to define the granularity. Otherwise the system could be gamed in such a way that you could cause an entire document to be classified by inserting an otherwise unrelated footnote containing sensitive info.
In my experience, most information is classified by virtue of it being aggregated and manipulated on a classified system. The only way to declassify that information is to have someone laboriously scrub all the data you want to put on an unclassified system and get it approved. if your entire network is classified, there's going to be a lot of mundane stuff there that is classified by default, because securing such a network requires that people can't just take data off of it without oversight.
I have worked on classified systems. We would have rules drawn up for what was classified and what was not. If you can draw up clear rules with no edge cases, no missed possibly-leaky information, where decisions are unambiguous and 20 people would always make exactly the same decision for every piece of information, you should go file a patent for that method because it is a truly novel invention. And we were only covering a single tiny domain of data, not trying to build a rules set for all possible data that could ever exist.
Of course, that's just the first step. Next we'd have to define who or what makes the determination for a given document.
And we'd have to define the granularity. Otherwise the system could be gamed in such a way that you could cause an entire document to be classified by inserting an otherwise unrelated footnote containing sensitive info.