> Applications is a wide field, but I'd be interested in specific examples where you think it wouldn't work.

Sure. Take a game with voice chat. Child mode disables voice chat. How does the game, which presumably uses a load of telemetry, avoid incidentally leaking which users are children via the lack of voice telemetry data coming from the client? It's probably possible, but the fact is we're talking about third party code running on a computer, and the computer running different code paths based on some value. The third party code knows that value, and if it has internet access can exfiltrate it. In that sense, if there's an internet connection, there's not a meaningful difference between "the OS tells the service/app your age rating preference" and "the OS changes what it displays based on your age rating preference."

Though while I'm throwing out fantasy policies we could solve this by banning pervasive surveillance outright.

You're assuming that everything not mandatory is prohibited. If the device is required to provide every service with the flag, every service gets the flag, even if it contains no adult content or adult content that the user agent could display or not without the service having a way to know about it.

The service would then have to deduce the information instead of getting it explicitly and may be able to do that some of the time instead of all of the time, which is an improvement. And then people can work on anti-fingerprinting technologies with the premise that if they succeed it actually does something, instead of the information being required by law to leak to the service.