That doesn't work because a) not all devices have cameras or trusted whatever and b) there are legitimate reasons to be able to use your ID without being there.

Suppose I want my search engine to be able to index the content on adult sites so it can return those results to adult users. I have a valid ID so I give it to my web crawler, that doesn't mean I'm willing to sit there all day and have it scan my face every 50ms when it moves on to the next website.

> And perhaps you do randomized (risk-based) audits where you actually have to call the government to check in

So then Bob calls in and says "Hi, this is Bob, yes this is a valid ID" and then carries on using it to supply age verification to anyone who wants it.

Unless you mean you're going to actually look into what Bob is using his ID to do, but now we're back to invasion of privacy, aren't we?

> - but in general, physical ID also has the same high-level problem of some baseline of forgeries, and the name of the game is just making it expensive, not making it impossible.

The difference is that physical forgeries are per-individual and generally require the acquirer to meet with the seller in person and risk criminal penalties, whereas digital things can be provided anonymously over the internet with no unit cost.

> One example of “make it expensive” would be to require unique device IDs to be registered, eg you bring in your iPhone (or Yubikey or whatever) and the DMV verified it’s not actually a non-certified device. This rests on keys being expensive to extract from the Secure Enclave.

Bob plugs his certified Yubikey or whatever into his server and has it age verify whatever anybody asks it to over the internet. The key never leaves the device.

Also, even if that wasn't the case, you would only need one person to extract a key from one device. This is the same reason DRM has failed to prevent Hollywood movies from appearing on piracy websites even if the median pirate doesn't know how to extract AACS keys or decrypt Netflix streams. (The reason they persist with the charade is that it prevents third parties from presenting a unified interface across media types and streaming services without their approval, not because it has ever been useful against copyright infringement. How many devices can recommend Netflix shows or YouTube videos based on what you watched on Disney+?)

The problem in this case is even worse because the "vulnerability" is actually an intrinsically mandatory feature. If you have a system that can't allow Bob to age verify whatever he wants to without revealing to the site who he is or the government what he's doing, your system is unacceptable. If you have a system that can do that and then Bob wants to age verify everything for everybody, your system is pointless.