Ethereum Institutional · Research · Quantum

The two algorithms

Only two quantum algorithms matter for cryptography, and they do very different things: one breaks public-key signatures, the other only weakens hashes. A third threat, from machine-generated mathematics rather than quantum hardware, points to the same fix.

The two algorithms

Shor breaks public-key signatures

A large enough quantum computer recovers a private key from its public key. That covers every elliptic-curve scheme a ledger uses: transaction signatures, validator signatures, and the pairings behind data commitments and most SNARKs.

More

No such machine exists yet; the migration must be done before one does. The estimated cost keeps falling. In March 2026 Google Quantum AI researchers and co-authors estimated fewer than 1,200 logical qubits, running in minutes on fewer than half a million physical qubits, and published a zero-knowledge proof that their circuit exists rather than the circuit itself, after consulting the US government; an independent researcher rediscovered the main optimization by June. By July an open challenge in which researchers worked with AI agents (ECDSA.fail) had published a circuit for the core step at more than 50% lower cost on Google’s own measure of qubits times gates, though the two count differently. The public record of attack costs may therefore lag what some parties know.

NIST’s post-quantum standards (August 2024) →

Sources (4)

Babbush, Zalcman, Gidney and others, Securing Elliptic Curve Cryptocurrencies against Quantum Vulnerabilities: Resource Estimates and Mitigations (arXiv 2603.28846, 30 March 2026, revised 15 April 2026): fewer than 1,200 logical qubits and 90 million Toffoli gates, or 1,450 and 70 million; minutes on fewer than half a million physical qubits; circuits withheld, validated by a zero-knowledge proof (checked 2026-11-14); Eigen Labs, From a hidden quantum circuit to open autoresearch: the ECDSA.fail story (10 September 2026): Google disclosed thresholds and a zero-knowledge proof, not the circuit; an open effort of more than 100 participants and their agents reached 1,151 logical qubits and about 1.3 million Toffoli gates for one point addition, more than 50% below Google’s reported qubit-times-Toffoli figure, under different accounting conventions (checked 2026-11-14); ECDSA.Fail: Open Autoresearch for Optimizing Elliptic-Curve Point Addition in Shor’s Algorithm (arXiv 2609.09582, September 2026): the challenge launched in late May 2026; individual point additions, not a complete Shor computation (checked 2026-11-14); Bankless, Justin Drake puts quantum Q-Day odds at 50% by 2032 (2 June 2026; secondary, reporting his post on X): 10% by 2030 and 50% by 2032; Google kept its optimizations behind a zero-knowledge proof after engaging with the US government, and an independent researcher rediscovered the main one (checked 2026-11-14)

Grover only weakens hashes

It halves a hash’s effective strength, so 256-bit hashes behave like 128-bit ones: weakened, not broken. Addresses, state trees, hash-based signatures and STARK proofs all survive, which is why the fix is to rebuild on hashes.

ethereum.org on quantum resistance →

A second threat: cryptanalysis by AI

Since October 2026 Ethereum Foundation researchers have warned that advances in machine-generated mathematics could weaken the assumptions behind elliptic-curve signatures before any quantum computer does, in their worst case within months rather than years; no such attack on curves has been demonstrated, and the warning is a research view, not a finding.

More

The effort is now public: open challenges set researchers and their AI agents on classical attacks against the curves Ethereum uses, BLS12-381 among them, with results accepted as exploratory rather than as proofs of a break, and none has broken a curve. On lattices there is now a demonstration: in July 2026 an AI model found an unexploited symmetry in HAWK, a third-round candidate in NIST's additional signature process, roughly halving its effective key size, and the HAWK team withdrew the scheme the next day; the attack stays exponential and is specific to HAWK, and nothing deployed changed. The reasoning is that curves have rich structure for a clever attack to use and hashes have as little as cryptography offers, which is why the same fix serves both threats: build on hashes, and verify the implementations formally, which has become far cheaper. Ethereum Institutional reads it as a reason to prefer the schemes with the fewest assumptions and to keep signature schemes replaceable, not as a forecast.

Ethereum researcher says AI may break encryption before quantum does (Decrypt, 7 October 2026) →

Sources (6)

Decrypt, Ethereum researcher says AI may break encryption before quantum does (7 October 2026; secondary): the worst case in months not years; bunker mode; curves vulnerable, hashes preferred; no attack demonstrated (checked 2026-11-06); CurveSmash (Yukon, hosted by Eigen Labs): open cryptanalysis challenges on secp256k1, P-256, BLS12-381 and BN254, scored in log2 of work; acceptance is exploratory, not proof or a practical break (checked 2026-11-14); Anthropic, Discovering cryptographic weaknesses with Claude (28 July 2026): an unexploited automorphism in HAWK's lattice, key recovery for HAWK-256 demonstrated at about 2³⁸ operations against an estimated 2⁶⁴, the attack still exponential and specific to HAWK; a 200 to 800 times faster attack on seven-round AES-128, still impractical; no production software has to change (checked 2026-11-10); NIST, Nine candidates advance to the third round of additional digital signatures (14 May 2026, updated 29 July 2026: the HAWK team withdrew its algorithm on 29 July 2026 after the AI-found vulnerability) (checked 2026-11-10); Devcon India, post-quantum Q&A with Ethereum Foundation researcher Justin Drake (November 2026; video, transcript read 6 November 2026): earliest 2029 and expected 2032, three years earlier than two years before; coordination the bottleneck, one upgrade about every ten months, 2028 at the earliest on the current cadence; a lean adaptation of SPHINCS+ for accounts, hash-based over lattices for the L1; proof of seed; custodians in bunker mode; the post-AI threat (checked 2026-11-06); Ethereum Foundation, Post-Quantum Ethereum portal (updated 27 June 2026): not imminent, most roadmaps early to mid 2030s; Execution, Consensus and Data layers; leanXMSS, leanSig, leanVM, leanSpec; SPHINCS+ a candidate for the execution layer; cryptographic agility; L1 upgrades by 2029; a USD 20 million formal verification project (checked 2026-11-06)

Help shape this page. Open an issue to suggest a source, a deployment or a question your risk committee has asked.

This map is an Ethereum Institutional house view for discussion, not legal advice. © 2026 Ethereum Institutional. Licensed under CC BY-SA 4.0 (share and adapt with credit, under the same license). Version 1137e34 · built 2026-11-16