DEV Community

Cover image for What the Binance custody breach actually shows: valid signatures, concentrated authority, and the limits of proof of reserves
Alfio Musumeci
Alfio Musumeci

Posted on Originally published at alfiomus.blogspot.com

What the Binance custody breach actually shows: valid signatures, concentrated authority, and the limits of proof of reserves

#cryptocurrency #security #cryptography #ethereum

A technical, structured edit of the essay.

TL;DR

  • On 2026-10-29, an attacker drained Binance operational (hot) wallets across ~10 chains. Estimated loot: $6.3-6.43B depending on the source - the largest recorded theft of its kind.
  • Public evidence points to compromised signing infrastructure, not broken cryptography: valid signatures were produced. On Ethereum, the "Binance 14" wallet was delegated to an attacker contract via EIP-7702 and drained in one transaction; "Binance 16" signed ordinary transfers with its own key.
  • Affected wallets backed customer balances (they appeared in the Oct 1 proof of reserves). Withdrawals were suspended the same day.
  • As of Nov 1: Binance reports 95% of assets replenished from SAFU + futures insurance fund + corporate treasury (90% on Oct 31); a new PoR snapshot (Oct 31 00:00 UTC) is published; the full zk proof was announced for Nov 3 and was still pending; withdrawals began reopening in phases, announced by the company (first window Nov 1 06:00-09:00 UTC, a 15-network phase one starting with BTC, per-network security checks, no fixed date for the rest); independent on-chain trackers observed about $2B leaving Binance-labeled wallets in the first three hours, consistent with the reopening but not publicly matched to customer withdrawals.
  • Company's preliminary root-cause hypothesis: a compromised employee laptop. Still a company statement; a detailed incident report is promised after the investigation.

1. The technical vector (as evidenced so far)

No mathematical break is in evidence. The interesting part is authorization:

  • EIP-7702 delegation: an EOA can delegate its execution to a contract. One delegation turned a known exchange wallet into a shell for the attacker's contract - one transaction emptied it. Valid key, valid signature, malicious logic.
  • Ordinary signed transfers: a second wallet simply signed transfer transactions. That points at key/signing-path compromise, not protocol failure.

Operational takeaway: the perimeter that failed is the one that authorizes movement - endpoints, signing paths, delegation mechanics - not the elliptic curve.

2. What is publicly verifiable, and what is not

Verifiable without taking anyone's word:

  • Fund movements on public ledgers, reconstructed by independent researchers (Elliptic and others).
  • PoR snapshot existence and the listed reserve addresses; user-level inclusion checks once the full Merkle/zk proof is published.

Not verifiable that way:

  • Liabilities. The ledger shows assets, not debts; a PoR snapshot is a snapshot at a point in time, not a solvency audit.
  • The full authority structure. Public labels show wallets, not who can sign behind them or how many such points exist.
  • Root cause. Currently a declaration by the affected company.

3. The concentration question

Balances are distributed across millions of users; the capacity to move them sits in far fewer points. This attack needed no per-customer account breach - it reached the wallets where custody concentrates. Concentrated value is a concentrated target.

Concentration cuts both ways, though: the same point that concentrates value lets defense concentrate (better teams, better controls, more eyes on fewer doors). "Defenses are weaker than the attack" is a hypothesis, not a demonstrated fact. The open tension: how many such points exist, how identifiable they are, and whether single-failure damage grows faster than the defense around it. A system's security here is not only how many accounts it protects, but how much damage one failure allows.

4. What this episode does not show

  • It does not show broken cryptography (evidence points at custody/signing systems).
  • It does not show protocol failure (evidence points at the custodian; full exoneration would also be premature).
  • It says nothing about quantum computing. Two papers from 2026-03-30 (arXiv 2603.28846; arXiv 2603.28627) estimate resources for future quantum attacks on ECC - theoretical, machines that do not yet exist, not an occurred attack.

5. Historical rhyme (short version)

1933/1971: conversion promise rewritten by explicit rule change (political decision). 2001/2002 Argentina: withdrawal restricted, then terms changed (systemic collapse). 2026: third party steals custodied assets, custodian pauses withdrawals while recovering (criminal attack). Different causes, one repeated shape: the interruption is decided or suffered by whoever administers the promise, and holders face a repair they do not control. Five variables worth keeping separate: convertibility, liquidity, solvency, custody, issuance.

Facts verified 2026-11-01; the Binance situation is still moving. Questions, not prediction or investment advice.

Sources

Suggested tags: security, cryptography, ethereum, custody

Top comments (0)