What might those be? Everything I've seen is ineffective or can be compromised. I think letting parents parent is probably the best option, but will require better/easier/free tools and education to help lock down different categories of harm.

Website generates a random nounce, the government offers to sign the json { over_18: true, nounce: [inserted_here] }. That signing is coupled to you being signed in, or scanning your passport via nfc or whatever

That significantly reduces the leaked information: the service doesn't receive any information who you are, and the government doesn't know what service you use. They can collude, but that does not reveal more to them than most current age-verification methods. Children can get another adult to sign their request, but that's a working attack on nearly any age verification (including in-person purchases)

Of course this doesn't replace parenting

There's nothing that can't be worked around or compromised in some way. Even the face scans and ID scans being used today are being circumvented.

In the end we do need parents to parent, but websites should help. The easiest way would be for websites to send an HTTP header to the client so that censorware can know to block or allow it accordingly.