The Promise in Custody

The Promise in Custody

Three episodes decades apart and different in cause, joined by a single question: what stands behind a balance, and what happens to whoever depends on it when the promise breaks.

A balance is a promise

A banknote is, in itself, the money: it only promises that others will accept it. A bank deposit is something else: a bank's promise to hand over that money when you ask. A balance in an app promises something similar: that a company will move it at your request. The banknote rests on custom and law; the deposit and the balance are claims against someone, a number accompanied by the word of whoever administers it.

For thousands of years trade ran on coined metal, and value traveled inside the coin itself. The great transformation was not technical but verbal: at some point we started accepting promises instead of things. First promises of conversion (this paper can be exchanged for gold), then promises of acceptance (this paper is worth something because everyone accepts it), today promises of availability (this balance can be moved whenever you want).

When money lives in someone else's hands, it is not the thing of value: it is the promise that the thing will appear when you ask for it.

The paradox of one's word

There is a widespread intuition: everyday agreements are armored with paper (contracts, signatures, certifications), while a deal closed with a handshake sounds like another era. The great concentrations of wealth also have contracts, regulation and audits; they do not depend on anyone's word in a literal sense. And yet none of those safeguards removes the moment of trust: someone administers, someone audits, someone has to enforce. The question is not whether we trust, but in whom that trust came to rest and how much we can demand from them.

One possible reading of that contrast: the word did not lose value, it became concentrated. No need to claim that promises were kept more often in the past, nor to write an elegy for lost trust. The question is enough: we armor small agreements with paper; what do we know about those who administer the large ones?

Three times the same question

It is worth telling the three episodes with their differences intact. They are not the same event. They are three different answers to the same question: what happens when whoever administers the promise decides, suffers or causes its interruption?

1933 and 1971: the conversion rewritten

Convertible banknotes were a promise of metal: silver and gold certificates, receipts of a real deposit, and until 1933 several notes could be redeemed for gold. The break came in stages: in 1933 the State suspended redemption for citizens and ordered monetary gold delivered to the Treasury, paid at the prevailing price of 20.67 dollars per ounce, with exceptions such as small holdings, collectible coins and industrial uses. In 1934 it set the new official price at 35. Whoever had handed it in at the old price had no way to buy back: they absorbed the difference. Bretton Woods, in 1944, left redemption open only to governments. Nixon closed it entirely on August 15, 1971.

There was no theft and no bankruptcy: the issuer replaced the contract through an explicit change of rules. And purely fiduciary money was not born there either: China was already issuing inconvertible paper in the 11th century.

Sources: Federal Reserve History, Roosevelt's gold program (https://www.federalreservehistory.org/essays/roosevelts-gold-program) · Gold Reserve Act (https://www.federalreservehistory.org/essays/gold-reserve-act) · The end of convertibility (https://www.federalreservehistory.org/essays/gold-convertibility-ends) · Hoover Institution, paper money in imperial China (https://www.hoover.org/research/rise-and-demise-paper-money-imperial-china)

2001 and 2002: the wait and the change of terms

Argentina promised dollars under a rule that kept it from issuing them: one-to-one convertibility. When the run came, the December 2001 "corralito" limited cash withdrawals: the balance still existed and could move inside the banking system, but could not freely turn into banknotes. Weeks later, the "corralón" froze most deposits, and the February 2002 pesification changed what would be returned: dollars converted back at 1.40 pesos plus adjustment, while many debts remained on other terms.

First the imposed wait, then the asymmetric restructuring. Depositors bore part of the adjustment without having chosen it: they financed, through their immobility and their loss, the restructuring of a system to which they were creditors.

Sources: Decree 1570/2001 (https://www.argentina.gob.ar/normativa/nacional/decreto-1570-2001-70355/texto) · Decree 214/2002 (https://www.argentina.gob.ar/normativa/nacional/decreto-214-2002-72017/texto) · Brookings, the 2001 Argentine crisis (https://www.brookings.edu/wp-content/uploads/2016/06/11_argentina_kiguel.pdf)

2026: custody breached

On October 29, 2026 an attacker emptied operational Binance wallets across about ten chains. The haul, estimated at between 6.3 and 6.43 billion dollars depending on the source, makes it the largest recorded theft of its kind. Public evidence suggests the cryptography was not broken: the infrastructure that authorizes movements was compromised and valid signatures were obtained. On Ethereum, a wallet known as Binance 14 was delegated to an attacker contract through EIP-7702 and emptied in a single transaction; another, Binance 16, signed ordinary transfers with its own key.

Those wallets backed customer balances: they appeared in the October 1 proof of reserves. Withdrawals were suspended the same day as the attack. Binance says it has replenished 95% of the assets as of November 1 using its emergency fund, insurance and treasury (it reported 90% on October 31), and that it took a new proof-of-reserves snapshot, already published, with the ten main assets fully covered; the full cryptographic proof, announced for November 3, is still pending. The phased withdrawal reopening began on November 1, announced by the company: a first phase covering fifteen networks started with bitcoin as the first named batch, with progressive per-network security checks and no fixed date for the rest. Independent on-chain trackers observed about 2 billion dollars leaving Binance-labeled wallets in the first three hours, movements consistent with the reopening but not publicly matched to customer withdrawals.

Sources: Elliptic, independent on-chain analysis (https://www.elliptic.co/insights/binance-largest-hack-in-history/) · Binance, October 31 recovery update (https://www.binance.com/en/support/announcement/detail/1fdf1f61559c401da0435895ffcccb62) · CoinDesk, the phased withdrawal reopening (https://www.coindesk.com/markets/2026/11/01/binance-opens-crypto-withdrawals-in-phases-starting-with-bitcoin-following-usd6-3-billion-hack) · crypto.news, 95% coverage according to the company (https://crypto.news/binance-reopens-bitcoin-withdrawals-after-hack/) · Yahoo Finance, the outflows observed by Whale Alert (https://finance.yahoo.com/markets/crypto/articles/2-billion-moves-first-3-094627615.html)

The common path

In all three cases the same limited loss repeats: an ability the holder took for granted. In 1933 and 1971, converting paper into metal. In 2001, withdrawing the deposit in cash. In 2026, moving the balance off the platform. It is not the same mechanism: in the first case the rule changed for the holders affected by each measure; in the others, access was restricted while the balance remained on the books. But in every case the interruption was decided or suffered by whoever administers the promise, and those who depended on it were left subject to a restructuring they do not control: they wait, they accept new terms, they absorb costs.

The balance can keep existing; what disappears is an ability the holder believed they had.

The differences, firm and brief

1. 1933 and 1971: a political decision. The State rewrites the conversion rule through an explicit change of rules.
2. 2001 and 2002: a systemic collapse. A bank liquidity and solvency crisis under a rigid monetary rule.
3. 2026: a criminal attack. A third party steals assets in custody and the custodian pauses withdrawals while it recovers.

Five words not to conflate: convertibility (what the unit can be exchanged for), liquidity (whether it can pay now), solvency (whether assets cover debts), custody (who holds the asset) and issuance (who creates new units). These variables combine differently in each episode; treating them as one thing is the fastest way to misdiagnose.

Why believe the new promise?

The uncomfortable question is legitimate: why accept the explanation of whoever has just defaulted? Part of the story can be checked without taking anyone's word: the fund movements were recorded on public ledgers that independent researchers reconstruct and debate. Checking is not the same as auditing: reading those records takes expertise, and interpretations can also be wrong.

But transparency has precise limits. The ledger shows assets, not debts. A proof of reserves is a snapshot at a point in time; it does not audit full solvency. Tokens reappearing in a wallet does not prove that every customer can withdraw. And the root cause of the attack remains, for now, a statement by the affected company: its preliminary hypothesis points to a compromised employee laptop. The difference from the past is not that trust is no longer needed: it is that a portion of the promise became observable. Part of the announced evidence has already arrived: the reopening began and outflows are already observable on-chain, and the new reserve snapshot is public; the full cryptographic proof is not. Each milestone met provides further evidence, though not a full audit.

What this episode does not prove

1. It does not prove the cryptography was broken. Available evidence points to the custody and signing systems, not to a mathematical break.
2. It does not prove the asset failed. Available evidence points to the custodian and shows no protocol failure, though it does not fully exonerate it either. And the reverse symmetry holds: a working protocol does not prove the asset is a good investment.
3. It does not prove there is no alternative. Self-custody and decentralized finance exist, but they relocate the same problem: they change who answers for keys, errors and recovery, with schemes such as multisig or social recovery that distribute that burden in different ways.
4. It proves nothing about quantum computing. Two papers from March 30, 2026 estimated the resources a future quantum attack would need against elliptic-curve cryptography; they describe a theoretical vulnerability of machines that do not yet exist, not an attack that happened. That there is no evidence of a quantum vector does not make it impossible, but it does not put it on the same level as an evidence-backed explanation either.

Sources for the quantum point: arXiv 2603.28846, Google Quantum AI and Ethereum Foundation (https://arxiv.org/abs/2603.28846) · arXiv 2603.28627, Caltech and Oratomic (https://arxiv.org/abs/2603.28627)

Closing: where risk concentrates

We usually picture the great digital heist as millions of accounts breached one by one. This attack suggests the reverse path: there was no need to breach one account per customer; it was enough to reach the wallets where custody concentrates. Balances are distributed among millions of people; the capacity to move them, in far fewer points. How many there are and who can sign at each one is not visible on-chain: public labels show wallets, not the authority structure behind them.

That raises a strategic question, distinct from the technical one. The quantum-computing papers describe a possible vulnerability of the cryptography: what machines it would take to break it. But this attack suggests another layer, one that depends on no future machine: an attacker can concentrate resources on the few places where a single failure reaches millions of people at once. Concentrated value is also a concentrated target.

Concentration, however, cuts both ways. The same point that concentrates value allows the defense to concentrate too: better teams, better controls, more eyes on fewer doors. That defenses are weaker than the attack is a hypothesis, not a proven fact. What is visible is the tension: we do not yet know how many of those points exist, how identifiable they are, or whether the damage of one failure grows faster than the defense around it. The security of such a system is measured not only in how many accounts it protects, but in how much damage a single failure allows.

And there the opening question returns, at another scale. Trust is not the system's defect: it is its condition. Every attempt to eliminate it moves it: from metal to the issuer, from issuer to custodian, from custodian to code, from code to whoever writes and administers it. Perhaps the lesson is not to distrust everything, but to know where trust came to rest this time.

The question left open: if trust is unavoidable, what should a person be able to verify before granting it again?

--- Facts verified on November 1, 2026; the Binance situation is still moving: the withdrawal reopening is advancing in phases and the full cryptographic proof of reserves was still pending at the time of writing. This text raises questions; it is not a prediction or investment advice.