Shielded Labs

Introducing Epoch, Shielded Labs’ new research and engineering project focused on hardening Zcash from future threats:

AI-powered hackers
Quantum Computing risk
Adversarial governments

We plan to achieve this through formal verification of critical code and by deploying “post-everything” cryptography.

Epoch will ensure Zcash remains designed for the threats we face today and for whatever comes next. ⤵️

Zcash has a new browser based wallet: Zillions @zillions_app

If you are interested in learning what makes Zillions different, check out this article.

x.com/zillions_app/s…

Secure hash functions are a critical linchpin of digital infrastructure. Lately, Shielded Labs has been focused on strengthening them. Hashsmash is a collaborative effort to probe them, and Project Epoch has future-proof hash functions as a pillar of its focus.

Now BLAKE3-Servil has been released as a new leap forward. It is fast, like 10× as fast or more than competitors in many cases, and also secure: assembly-language parts come with correctness proofs, and the rest uses mutation testing.

BLAKE3-Servil achieves this because it was completely written by AI. Looking forward, we expect an update that makes the correctness proofs and/or memory safety complete.

How does an idea actually become a Zcash network upgrade?

This post walks through the full path from proposal to mainnet.

Taylor Hornby (@DefuseSec) on the state of AI-assisted software audits:

Current models are great at auditing, and everyone needs to be doing it. The two biggest problems that still need solving are:

(a) models are not good at properly understanding an issue’s severity, so there is still a lot of manual work to validate what the AI finds (this is the next thing I want to work on), and

(b) defenders need access to the best models without cyber safeguards in order to audit their own code.

Frontier labs need to do a better job of letting legitimate security research and auditing through the safeguards, and of providing AI audits and patching support to open-source projects. Anthropic’s Fable safeguards’ initial refusals to work on pretty much anything cryptography-related may have ultimately done more harm than good, since they blocked legitimate defensive research.

The situation has gotten better now that Anthropic has allowed users in their Cyber Verification Program to run Mythos on security tasks.