Platforms can't know what the false positive and false negative rates are simply from cases where a user is thought to be a minor and proves otherwise. Even if the law were written better, for example to say that 95% accuracy is acceptable, platforms would still have to verify id from at least a random sampling of thought-to-be-adult users, which doesn't help you if you're one of the ones randomly selected.

We need some kind of verification system that gives no extra information about users to the platforms, but I don't know if there's a true ZK way; it might require government involvement. I think that's fine. Govts could certainly run an age-verification system, give a signed yes or no token back to the user, with some permanently applied jitter per person so that platforms can't use cookies from returning users to figure out their birthday. As long as the government program has strict oversight to ensure it's not saving information about who's visiting what sites, it seems fine, or at least vastly better than entrusting photos of IDs to private 3rd parties.