For example, imagine you put the same private key into the 'secure element' of every single iphone. You use code signing so that key is only unlocked when the phone is running unmodified iOS with all security updates. You use encryption and remote attestation for the front-facing camera and face id depth sensor. You use NFC to read government-authenticated age and appearance data from biometric passport chips (or digital ID cards) and you store it on-device.
Then, when you want to access pornhub, they send an age challenge to your device, your device makes sure your face matches the stored passport, and if so it signs the challenge with the private key.
Pornhub gets an Apple-signed attestation of age - but because every phone signs with challenges with the same private key, Pornhub can't link it to a particular phone or identity document.
So in a very narrow sense, privacy is preserved.
You can't use someone else's ID, as it checks your face every time. You can't fool it with a photo of the person because of the depth sensor. You can't MITM/replay the camera/depth data because the link is encrypted. You can't substitute software that skips the check with a rooted phone because of the code signing. Security holes can be closed by just pushing a mandatory OS update.
Sure, it doesn't work on PCs. Doesn't work on Linux, or on unlocked/rooted phones. It hands users' government ID documents over to Google and Apple. It requires people to carry foreign-made, battery powered, network connected GPS trackers (with cameras, microphones and speech recognition) with them. And there are non-negotiable terms of service everyone must agree to. But if you define "privacy-preserving" to ignore all that stuff and only consider whether Pornhub learns your identity, it's privacy-preserving.
Also even if it doesn't get leaked directly, the security of TPM chips is not absolute. Secrets from them can theoretically be extracted given an attacker with sufficient means and motivation. Normally nothing that's on a typical TPM chip would warrant a project of that magnitude, but a widely used private key can change that equation.
Plus a TPM chip doesn't really have means to tell the phone isn't being lied to. You could swap out the actual phone camera hardware and sensors for a custom board that feeds the entire phone camera data of your choosing and it would be none-the-wiser.
There's not necessarily wrong. Despite the vapid and damaging nature of most popular online media, isolating a child from it might have even worse social consequences when their real-life peer groups discover that they're not on social media or that their parents have neutered their phone. Some kids would turn out fine after that. Others would be socially destroyed for life (maybe with the right therapy they could become well-adjusted, but high quality therapy is rare).
No, they are a solution for parents who want to use them, and that's all they should be. Their existence demonstrates that it's possible to handle this without regulation, other than the desire of some people to inflict their preferences onto other people's kids.
Do they work currently? Not really
Are they too complex for the avg joe to work out. Unfortunately yes. (Something about the smartest bears and the dumbest humans)
The problem is that while you might be able to trust the crypto, the government won't trust you to do the crypto entirely by yourself. And this introduces avenues for deanonymisation. Moreover, collusion between the government and the entity making the age check can also theoretically deanonimize.
It's a complicated problem.
We continue to seek a technological solution to a parenting problem.
Then it's technically possible (and really not that difficult) for states to provide a service that issues zero-knowledge proofs of facts like "age > X".
> Unlinkability is achieved by design through Zero-Knowledge Proof cryptography see the "Privacy by design" section below.