Justin Drake: AI, not quantum, may be first to break crypto's signatures

Ethereum researcher Justin Drake argued in October 2026 that artificial intelligence, not quantum computers, could be the first to break the elliptic-curve cryptography that secures Bitcoin and Ethereum. Alex Pruden, co-founder and CEO of Project Eleven, and Conor Deegan, its co-founder and CTO, examined the claim with host Laura Shin on Unchained's How Crypto Can Survive If AI or Quantum Computers Break Its Cryptography, published 21 October 2026. The question matters because a break that arrives through AI would not come with the years of warning the quantum timeline is assumed to give.

The argument

Shin read Drake's post on the show. Drake wrote that "it is now reasonable to brace for the possibility that ECDSA breaks before Q-day. In the worst case, in months, not years" (from 48:15), and defined a break as "fast private key recovery, e.g., in one week on available hardware, e.g., a large GPU cluster." (from 48:15). His reasoning, as Shin read it, is that elliptic curves carry rich algebraic structure, while hash functions do not.

Deegan opened with a disclaimer: no known attack breaks elliptic curves today. But he said post-quantum schemes built on algebraic structure had already shown the risk: "We've seen two schemes, one called Hawk, which was built off lattices, another called McEliece, which is a different type of crypto system." (from 10:47). He then argued the public fight over which scheme to adopt misses most of the work: "I think the issue is it's kind of the wrong question. You know, which cryptographic scheme or which hardened system you're going to go for is fine, but it's the tip of the iceberg." (from 39:31).

What it rests on

Half of Deegan's evidence checks out. In July 2026 Anthropic said its Claude Mythos Preview model found an improved attack on HAWK, a lattice-based signature candidate still under NIST evaluation, that cut the scheme's security margin by reducing its effective key size. The work took about 60 hours and roughly $100,000 in API usage. HAWK was never standardised or deployed.

The McEliece result is different. The original paper describes a classical, quasipolynomial-time distinguisher for Classic McEliece that came out of research on private information retrieval, not an AI system. A follow-up that cut key-recovery costs further came from an Anthropic researcher; neither paper credits an AI model with the attack. Its authors call the improved attack estimates "not yet practical". McEliece is also an encryption scheme, not a signature scheme.

Pruden added a point that cuts against simple answers. NIST has standardised two post-quantum signature schemes, a lattice one and a hash-based one, and its primary recommendation, the lattice-based ML-DSA, is the one with more algebraic structure: the property Drake's argument treats as the weakness. Pruden also pointed to custody, using Fireblocks as the example: the signature scheme is one part of a system that has to change as a whole. The hash-based alternative, SLH-DSA, comes with larger signatures and slower performance.

The guests also separated signatures from encryption. Hash-based signatures can replace vulnerable signature schemes, but there is no hash-based way to agree a shared secret, so encryption and privacy protocols have no equally simple fallback. Johns Hopkins cryptographer Matthew Green had written on 9 October 2026 that he saw a 15% chance "we functionally lose confidence in our existing public-key encryption algorithms".

Ethereum already has a migration proposal on the table. EIP-8450, a draft "Post-Quantum Validator Key Registry" created on 15 October 2026, would have validators register dormant hash-based keys ahead of a move away from BLS signatures. ETH Daily mentioned it on 22 October 2026 as a separate draft proposal to have validators switch to post-quantum secure signatures. Drake is one of its four authors.

Where it was discussed

• Unchained: How Crypto Can Survive If AI or Quantum Computers Break Its Cryptography, 21 October 2026, from 8:35
• ETH Daily: Glamsterdam Hoodi Upgrade Postponed, 22 October 2026, from 1:52