Skip to content
Crypto.world

Research

Correctable Enrollment and One-Transaction Seed Proofs for Ethereum's Post-Quantum Transition

A concrete Ethereum migration path for quantum-exposed accounts, with released code, scheduled ECDSA retirement, and BIP-39 seed-recovery proofs.

arXiv
2611.25663
Published
2026-11-19
Authors
Xin Jin, Maximillian Liao, Handong Chen, Yi Nancy Li, Junyu Liu

AI summary

Overview

Research area: Blockchain security and cryptography — specifically, migrating Ethereum's account and validator keys from elliptic-curve signatures (ECDSA over secp256k1, and BLS over BLS12-381) to post-quantum alternatives before a cryptographically relevant quantum computer (CRQC) can extract private keys.

Technical level: Advanced. The paper assumes familiarity with Ethereum's execution and consensus layers, BIP-39/BIP-32/BIP-44 key derivation, zero-knowledge STARKs, and threshold finality.

Scope (one sentence): The paper designs, proves, implements, and benchmarks at mainnet scale a two-stage Ethereum migration in which owners enroll post-quantum keys cheaply with existing ECDSA signatures, any such enrollment remains correctable by a zero-knowledge proof of the wallet's BIP-39 seed until a fixed deadline, and validators switch to hash-based keys first.

What This Paper Is About

Any Ethereum account that has sent a transaction has published its public key, and a quantum computer running Shor's algorithm could derive its private key from that public key — such exposed accounts hold 54.6% of execution-layer ETH. Because the first such machine may be kept secret, a migration cannot wait for visible thefts, and once keys are extractable, legacy signatures can no longer distinguish an owner from an attacker. The paper's goal is a live-chain migration protocol that works without knowing when the quantum threat begins: cheap ECDSA-authorized enrollment before a scheduled cutoff, plus a compact zero-knowledge seed proof that can both rescue frozen accounts and override fraudulent enrollments.

Key Contributions

  1. A security audit of what a quantum computer breaks in Ethereum (Sections II, VI, and Appendix E). The authors map the affected surfaces (Table XIV) and prove that any migration of this kind needs two upgrades: an owner check based on an unexposed secret from which the account derives, and timely finalization of new validator keys (Proposition 2).

  2. A correctable migration protocol (Sections II, III, VI). Owners enroll new keys cheaply, and a seed proof can correct any enrollment — including a stealth enrollment by the adversary — until a common deadline T + W, with W set to 365 days. The account guarantees are defined as an "authenticated migration" game, proven with concrete bounds under stated assumptions, and model-checked.

  3. A seed proof that fits in one transaction (Section V). A three-layer recursive hash-based STARK covering Ethereum's full key derivation — six HMAC-SHA512 evaluations along a five-level BIP-44 path, three secp256k1 scalar multiplications, and Keccak-256 — produces a proof of about 190 KB that verifies in 38 ms.

  4. Validators switch first (Sections IV, VI). Validators register hash-based leanXMSS keys with their BLS keys before T₀ and the chain moves to Lean consensus from a finalized checkpoint at T₀, so recoveries can still reach finality under a stated bound on the attacker's voting power.

  5. Cost at mainnet scale (Section VII). A ten-node simulation with mainnet-derived accounts and EIP-1559 fee demand shows that block space, not proving, sets the migration's duration.

Main Findings

  • Exposure is large. Transaction-exposed externally owned accounts (EOAs) hold 54.6% of execution-layer ETH, and 61% of staked ETH returns to such EOAs when withdrawn. Of the 76.7 M ETH on the execution layer and 44.2 M ETH of stake (mainnet data from 3–4 November 2026), a majority sits behind published public keys.

  • The quantum cost of key extraction has fallen sharply. The estimated gate count to recover a single 256-bit key has dropped by three orders of magnitude since 2017, to 7 × 10⁷ Toffoli gates on fewer than 1,450 logical qubits — within reach, in physical error-corrected hardware, of a trapped-ion machine with about 20,000 physical qubits running for 26 days or a faster superconducting machine with under half a million physical qubits running in minutes.

  • The seed proof is small enough for one transaction. About 190 KB, verifying in 38 ms inside go-ethereum, with a median generation time of 111 s and 31 GiB of memory on a laptop CPU. It is nearly 30× smaller and 60× faster to verify than the recent EdDSA seed proof, which measures 5.4 MB and takes 2.3 s to verify. Four proofs aggregate into one aggregate proof of about 190 KB, so one batch transaction migrates four accounts.

  • Block space, not proving, sets migration duration. For the 4.45 M EOAs holding at least 0.1 ETH, migration takes 29.4 days if all enroll before T, 235 days if all recover after T in batches of four, and 165 days on a mixed path where accounts active since November 2023 enroll before T and the rest recover in batches. At a 1 gwei bid, migration transactions take 40% of the 30 M gas target.

  • Mempool exposure is the binding constraint for enrollment. During a mass enrollment, an ECDSA enrollment waits a median of 19.1 minutes in the public mempool — longer than the 9 to 12 minutes estimated for a fast superconducting quantum computer with precomputation to extract its private key. The authors therefore recommend that owners holding a seed and an unexposed key migrate by seed proof rather than enroll.

  • Fee costs are quantified. A self-paid recovery consumes 8.74 M gas (about half the 2²⁴ per-transaction gas cap); a batched four-account migration consumes 15.07 M gas. At the 1 gwei cap, an enrollment costs 0.00057 ETH and a batched recovery 0.00416 ETH per account; the mixed path costs the 4.45 M EOAs 13,100 ETH in total fees. As a cost reference, a rented A100 machine proves for $22 per thousand owners.

  • Validator recovery is threatened by aggregation. The 7,647 largest validators hold more than a third of mainnet stake; recovering their BLS keys one by one is estimated at 48 to 122 days on one superconducting quantum computer, but a single discrete logarithm of the sum of their public keys forges the slashing evidence for all of them. This motivates switching consensus first.

  • Centralized exchanges are a gap. Exchanges hold at least 14.6% of execution-layer ETH, and none of the 25 examined documents a mnemonic seed, placing their accounts in the "owners without a seed" group. Their 1,869 EOAs holding at least 0.1 ETH could finish enrolling at T₀ in 17.7 minutes of block space for 1.06 ETH in total fees.

  • Soundness is modeled, not assumed. Each proof layer has 128 to 130 bits of modeled round-by-round IOP soundness; quantum zero knowledge and weak simulation extractability of the composed proof remain explicit assumptions.

Methodology in Plain English

The authors split the problem into two stages. In the first stage, a "PQ-ready fork" at time T₀ turns on post-quantum transaction types while ordinary ECDSA signatures still work, opening an enrollment window [T₀, T). During this window an owner can attach a post-quantum key (such as ML-DSA-44) to their account with a cheap ECDSA-signed transaction (type 0x7d, 175k gas) and a proof of possession of the new key. At the second fork, T = Q_p, ECDSA authority is permanently retired; unenrolled accounts are frozen rather than destroyed.

Because an attacker with a quantum computer could forge an enrollment during that window, every enrollment stays correctable until T + W. The correction is a transaction (type 0x7e) carrying a zero-knowledge proof that the sender knows the BIP-39 seed from which the account's key derives — a secret that extracting a private key does not reveal, since inverting the initial HMAC-SHA512 admits only generic quantum search. The proof's public input binds the account, a migration nonce, the new key commitment, an expiry, and a fee cap, so a sponsor who pays the gas or a builder who reorders the block can neither substitute the key, replay the proof, nor overcharge.

To make the proof small, the authors build a three-layer recursive STARK from hash functions alone, with two layers of recursive compression and hiding commitments added at the outer layer for zero knowledge. A two-layer recursive aggregator combines four owners' final proofs into one.

Validators move first: before T₀ each validator uses its BLS key to register hash-based leanXMSS keys, and at T₀ the chain transitions to Lean consensus from a finalized checkpoint, excluding validators with conflicting registrations. This rests on an explicit capacity bound — before T₀ the adversary cannot forge BLS authorizations for a consensus threshold, and validators whose keys it registered hold less than a third of the new set.

The design is implemented in go-ethereum paired with the Ream Lean consensus client, the security goals are stated as a game and proven with concrete bounds under assumptions A1–A8, the rules are model-checked, and the full system is evaluated on ten nodes with mainnet's account population and ordinary EIP-1559 demand.

Why This Matters

Impact on research. The paper reframes post-quantum migration as a correctability problem rather than a detection problem: since the arrival of a CRQC is unobservable, the protocol is built so that its security does not depend on knowing when the threat begins. It also sets a concrete engineering bar for post-quantum zero-knowledge proofs on Ethereum — one transaction, 190 KB, 38 ms — against a prior 5.4 MB, 2.3 s baseline, and it quantifies live-chain migration throughput under real fee competition rather than assuming a fixed block-space share.

Real-world applications:

  • Wallet migration UX: wallets can enroll keys with an ordinary signature before the cutoff, or skip enrollment and migrate by seed proof, with derivation tied to the existing mnemonic so no new backup is needed.
  • Custodian and exchange planning: the analysis of exchanges' 1,869 large EOAs and their 17.7 minutes of block space gives custodians a concrete scheduling target.
  • Validator operations: the leanXMSS registration and checkpoint-exclusion procedure gives staking operators a concrete pre-T₀ checklist, including exits for hijacked validators with protected withdrawal accounts.
  • Smart contract triage: the finding that contracts accepting only recovered secp256k1 signatures — holding at least 4.8% of the stake that withdraws to contracts — freeze at T tells operators which contracts must be replaced beforehand, and the ERC-1271 isValidSignature path (provisional 50,000 gas) tells them how permits can keep working.

Industry relevance. The work targets a live, high-value network with hundreds of thousands of validators and millions of accounts, and it engages directly with existing proposals — voluntary "off-ramp" schemes and Buterin's emergency-hard-fork approach — arguing that the former fails once the adversary arrives and the latter is costly because it rolls back finalized history. Its conclusion that mempool wait time (19.1 minutes) can exceed key-extraction time (9 to 12 minutes) is an operational warning for every ECDSA-signed migration on any chain, not just Ethereum.

Future Directions

  • Trusted-execution-environment proving. The authors name proving inside a TEE as future work, noting it would require trust in the hardware plus post-quantum attestation — a way to avoid shipping the seed to a hosted prover.
  • Handling owners without a seed. Proposition 1 states that once the adversary holds an account's key and the chain records none of the owner's keys, no protocol checking only that key can guarantee a binding; how to serve HSM-, MPC-, and non-standard-path wallets (such as Ledger's legacy four-level path) remains open.
  • Governance and deadline mechanics. The paper leaves the emergency fork that may move T forward to T′ < T as a policy tool, constrained by the window lemma (W′ ≥ T + W − T′); how such forks are triggered and coordinated in practice is not resolved.
  • Beyond the base layer. Layer-2 chains and bridges are explicitly excluded from scope, and contracts that accept only ecrecover signatures must be replaced rather than migrated — both are open problems for the wider ecosystem.

Target Audience

Ethereum core protocol developers and client engineers (go-ethereum, consensus clients), cryptographic researchers working on post-quantum signatures and succinct proofs, wallet and custodian security architects planning key migration, and staking operators who need to understand the validator transition. It is also relevant to standards and governance bodies who must schedule the forks T₀ and T and set the window W. Readers without a background in elliptic-curve cryptography, BIP-32/39 derivation, and STARK soundness will find the proofs and gas-level protocol details demanding.

Authors’ abstract

A quantum computer running Shor's algorithm can extract the private key of every Ethereum account whose public key is published, and accounts exposed by their own transactions hold 54.6% of execution-layer ETH. As the first such machine may stay secret, a migration cannot wait for visible thefts, and once keys are extractable, legacy signatures no longer tell owners from attackers. We design a migration in which owners enroll post-quantum keys cheaply with their existing…

The full abstract on arXiv
Read the original paper