Protocol & Mining
Self-custody and exchange custody, and who holds the keys
Protocol referenceSource: Bitcoin Developer Guide and BIP 32No venue is named or assessed; the comparison is structural.
Control is the ability to sign
On the Bitcoin network, ownership is not recorded anywhere. What exists is a set of outputs, each locked to a condition, and the ability to spend an output is the ability to satisfy that condition. In the common case the condition is a public key, and satisfying it means producing a signature with the matching private key. Whoever holds that key controls the coins. There is no registry to appeal to and no authority that can reassign them.
This is the fact that makes custody a meaningful distinction rather than a matter of convenience. When a person holds their own keys, they can construct and sign a transaction at any moment, without anyone's permission and without anyone's cooperation. When a third party holds the keys, that ability belongs to the third party. The customer's claim is not a claim on specific coins; it is a claim against the institution, enforceable in the institution's terms and in whatever legal system applies to it.
The distinction is easy to blur because both arrangements look similar from a screen. A balance shown in an exchange interface and a balance shown in a self-custody wallet are both numbers, and both move when a payment is made. What differs is what stands behind the number. In one case it is a key the user controls. In the other it is a promise.
What exchange custody involves
An exchange that holds customer bitcoin is running a custodial business. Customer deposits are pooled, and the exchange controls the keys to the pooled holdings. Internally it maintains a ledger of who is owed what, and when a customer withdraws, the exchange sends coins from its own holdings to an address the customer specifies. The customer never touches the keys that secure the pooled balance.
The counterparty exposure this creates is the same exposure any depositor has to any institution. If the exchange is solvent and operating normally, the arrangement works well: trading is fast, there is no key material to lose, and a forgotten password is a support ticket rather than a permanent loss. If the exchange becomes insolvent, is hacked, is frozen by a regulator, or simply halts withdrawals, the customer's claim is caught up in the failure. The bitcoin may still exist on the chain, but the customer cannot move it, because the customer never had the ability to move it.
It is worth being precise about what this does and does not say. It does not say that custodial arrangements are illegitimate or that every custodian will fail. Many operate for years without incident, and for active trading a custodial venue is often the only practical option. It says that the customer has taken on a credit exposure, and that the exposure is a real feature of the arrangement rather than a remote possibility. The relevant question for a holder is whether that exposure is one they have chosen knowingly.
What self-custody involves
Self-custody removes the counterparty and replaces it with a different set of risks. The holder generates a key, or a seed from which keys are derived, and is responsible for keeping it secret and recoverable. There is no institution to notify if it is lost and no one to reverse a transaction if it is stolen. The security of the holding is entirely a function of how well the key material is protected.
Those risks are real and they are not symmetric with the custodial ones. A custodian's failure is a shared event that affects many customers at once and tends to attract legal process and public attention. A self-custody failure is private, permanent and usually silent. A seed phrase photographed and stored in a cloud account, a key written on a note that is later discarded, a device that fails without a backup — each of these is a total loss with no recourse.
The compensating property is that self-custody has no counterparty to fail. There is no balance sheet to become insolvent, no withdrawal queue to join, no jurisdiction that can freeze the holding. The holder's ability to move the coins depends on nothing but the key and the network. For a holder whose main concern is the failure of an institution, that is the whole point of the arrangement.
The practical differences
The two arrangements differ in recovery, in privacy and in what happens at the edges. Recovery is the clearest contrast. A custodial account can be recovered through the institution's own process, which usually means identity verification and a support request. A self-custody holding can be recovered only from the seed phrase or backup, and if that is gone the coins are gone. The seed phrase is therefore not a password to be reset; it is the holding itself.
Privacy differs in a subtler way. A custodial venue sees the customer's identity, balances and transaction history, and is typically subject to reporting obligations. A self-custody holder transacts directly with the network, and the on-chain record is public but not tied to an identity by the protocol. Neither arrangement is private in an absolute sense, but the information each one exposes is different in kind.
In practice many holders use both. A custodial account for trading and a self-custody arrangement for long-term holding is a common split, and it is a reasonable one: it keeps the counterparty exposure proportional to the amount that needs to be liquid. The seed phrases and private keys page explains what the key material actually is, and the hardware and software wallets page covers how it is stored.
Sources and references
- Bitcoin Developer Guide, Wallets — key control, wallet types and the meaning of ownership.
- BIP 32, Hierarchical Deterministic Wallets — how a single seed produces the keys a wallet uses.
- Bitcoin Developer Guide, Transactions — the signing step that constitutes control.
Related reading
- Protocol, Transactions & MiningThe supply schedule, transaction mechanics and mining economics behind the price.
- Proof of WorkThe hash puzzle, the target and nonce, and why accumulated work secures the chain.
- Difficulty AdjustmentThe 2,016-block retarget, its caps, and the ten-minute target it defends.
- HashrateWhat hashrate measures, why it is estimated, and how it differs from difficulty.
- Mining PoolsPooled hash rate, share accounting, payout schemes and centralisation.
- Miner RevenueThe block subsidy plus fees, and how the mix changes across subsidy epochs.