USENIX Association

We are madly in the process of doing more patent filings, and while I welcome questions, some of the answers will be, "I'm sorry, but the patent for that hasn't gone in yet, and I can't talk about it." But I can talk about other aspects of the mill or more about this if you see fit.

Uh, coming up in USENIX, we're going to be talking about the, um, architecture of the machine, the basic category.

Remember when we started, there was a VLIW chip, and there was an out-of-order superscalar.

The mill is not one of those categories.

That's a new category, and it has, um, well, let me put it this way.

There are no general registers, and the programming model is quite different.

Um, and that will be presented at USENIX.

Coming up in September, we'll be talking about, uh, the way our memory systems-- subsystems works for the IEEE.

we're gonna figure that ahead of time we're gonna pursue it so i shall summon rob joyce so he ran offensive and defensive ops for nsa and he said in public he's written publicly about uh you know what makes nsa awesome at doing what they do is not as people often believe the fact they just have really great zero days and really great tools what makes them great is the fact that they take the time and they have the ability to deeply understand an environment they're operating in better than the people who designed it or are managing and maintaining at that time And that's what is concerning me.

He has another quote from like a Usenix paper where he says, like patience and persistence will get you places that a zero day never will.

Harnesses can do things when it comes to like patients.

It is like sucky for a human operator to sit on a machine and like dig through a bunch of files slowly.

Researchers who cataloged 141 security flaws in zero-knowledge proof systems between 2018 and 2024 found 99 of them in the circuits, the sets of mathematical constraints that decide whether a proof is accepted.

Almost 9 in 10 of the 141 broke soundness, the guarantee that a prover cannot convince a verifier of something false, according to a study presented at USENIX Security 2024.

A soundness bug does not crash a system or trigger an alert.

It lets a false statement pass as true.

It also concentrates risk.

Every program proved in Z-quasm passes through the same circuits, so a single missing constraint tweakens every proof built on top of it at once.

The USENIX study found that 95 of its 99 circuit bugs broke soundness.

Many came from constraints that were missing altogether or that translated the intended logic into constraints incorrectly.