Self-custody and exchange custody, compared honestly

When you deposit to a centralised exchange, the funds become the property of the operator and you hold a claim on their balance sheet. Your balance is an entry in their database. A withdrawal is a request they process, which means it can be delayed, restricted, or in the failure case not honoured at all. In return you get account recovery, a support desk, fiat rails, no gas costs and no key to lose, and for a large number of people those are worth more than the counterparty risk.

Funds sit at an address whose key you control. Moving them is a transaction that no operator approves and none can block. There is no counterparty to fail, no withdrawal queue and no account to freeze. The failure mode moves onto you: a lost key is unrecoverable, a signature approved carelessly is final, and there is nobody to reverse a mistake. Self-custody removes one class of risk entirely and hands you another.

The binary framing is out of date. In practice the arrangements available cover a spectrum.

• A margin contract you deposit into and can withdraw from without permission, which may still have an upgrade key
• Embedded wallets where the key is split into shares held between you, a provider and a backup
• Passkey-based wallets, where the secret lives in device hardware rather than in a phrase you write down
• Multisig and hardware signers, where one compromised device is not enough to move funds
• Session keys that let an app trade on your behalf within limits you set, without holding your funds

Rather than asking whether something is custodial, ask which specific action needs whose permission. Can you withdraw without the operator agreeing. If the website is offline tomorrow, is there a documented way to reach the contract directly. Who can upgrade that contract, is the key a multisig, and is there a timelock so you would see a change before it took effect. Where does the key material live, and what recovers it if the device is lost. The answers vary widely between products that describe themselves identically.

Exchange custody fails institutionally: insolvency, fraud, a regulator freezing withdrawals, a hack of the operator hot wallet. These events are rare, correlated across all users at once, and entirely outside your control. Self-custody fails individually: a phishing signature, a lost device, a malicious contract approval, a mistyped address. These are common, uncorrelated, and almost entirely within your control. Choosing between them is largely a judgement about which of those you are better placed to manage.

Keep trading capital separate from savings, whatever the model. Under self-custody, use a hardware or multisig setup for the balance you would not want to lose and a hot wallet for the amount you actively trade. Under exchange custody, treat any balance left on a venue as an unsecured loan to that venue and size it accordingly. In both cases, the thing that protects you is the amount you left exposed, not the strength of the opinion you hold about custody.