Homomorphic encryption and similar techniques in this paper are just getting going. They are impressive technologies. However, they often take 100x the compute of "regular" systems with encrypted networking. This is probably the main blocker for these types of technologies. Until and unless insurance companies mandate these technologies because they are tired of paying out for their customers getting hacked, they probably won't be deployed. Probably for the best. Most devs can barely make code without advanced math and encrypted data work, let alone these types of advanced platforms.

I agree that we have more capable+flexible cryptographic primitives than ever before, but I don't really buy the "Universal Protocol" thing.

For non-cryptographic uses we have "universal protocols" already, JSON being an example. You can adapt just about any format to and from JSON, if you want. But the fact that this is possible has not solved the interop problem, in the general case.

Similarly for "Hallucinated Servers". Even if you trust all nodes (and don't need cryptography), distributed computing is still kinda hard, and we have to write programs in particular ways to make them efficiently distributable. I'm sure this can work really well for some problem domains, but it's a subset.

I've been looking at the field, and I can't really see how most of this is useful. ZKPs and FHE add a lot of complexity to a pretty simple task: verifying the age and/or identity.

These tasks are so simple that you can _almost_ use the existing TLS client certificates for that. Their only drawback is that they're trackable. A simple asymmetric challenge-response system with a nonce easily fixes this:

1. The service provider generates a 128-bit nonce and sends it to me.

2. I use a verification system provided by my government, and it returns a document saying: "The owner is more than 18 years old, the nonce for the request was ......, and this proof is valid for this service name hash". This document is signed by the trusted government certificate.

3. I send this signed document to the service provider.

No need for range proofs and other stuff. I think this flow can even be expressed using OIDC and JWTs!

What am I missing that requires full-blown ZKPs?

Zk would perfect for online age verification, but governments do not want to implement it like this. Instead they want id and face collection for mass surveillance, using age verification as an excuse.

But won't this make the Palantir AI Overlord angery?