지케이싱크 ZK

박주혁27d ago

Vitalik's hot take on this: I don't want to hastily recommend moving funds to a new wallet right now. But we should take seriously the threat that AI poses to cryptography as it accelerates progress in mathematics. We need to minimize dependence not only on cryptographic techniques vulnerable to quantum computers, but also on cryptographic techniques that might be vulnerable to AI. From this perspective, the key areas where new risks are emerging are unfortunately ML-DSA, fully homomorphic encryption (FHE), and lattice-based cryptography. (This is also another reason why ECDSA might be broken faster than expected, in addition to quantum computers. This is why we recommend "new addresses.") So far, most people have thought: "Even if elliptic curves are broken, hashes and lattices are safe." But over the next two years, there's a significant possibility that the mathematical advances AI makes could seriously damage the practical security level of lattice-based cryptography. My basic scenario of concern is this: Integer factorization is a problem that takes 2^{n/2} time if approached naively. But over decades, brilliant people discovered and refined the number field sieve, reducing the time needed to 2^{O(n^{1/3})}. This is why RSA keys and signatures need to be about 400 bytes rather than 64 bytes. What if elliptic curves and lattices have similar hidden weaknesses? Weaknesses humans haven't been clever enough to discover yet, but that AI will soon find. The important reason Ethereum's Lean roadmap has moved toward "hash-only" over the past year is exactly this. Not using lattices, not ML-DSA, not Falcon, and not using lattice-based commitments even inside zero-knowledge proofs (ZK). The signatures and proofs Lean Ethereum adopts are all hash-based methods like WOTS and SPHINCS. We already know how to implement signatures and proofs using only hashes. The bigger challenge is public-key cryptography. This goes far beyond just blockchain. Public-key cryptography is needed for secure communications, anonymization protocols, and many other technologies. Unfortunately, mathematical theorems have long existed showing that public-key cryptography cannot be implemented with just hashes. We need a mathematical device called a "trapdoor" that makes it easy to invert if you know the secret, and this requires at least one usable "structure." This could be a structure based on group theory (isogenies

1.2KEthereumShow Korean original