Custodial vs Non-Custodial Wallets

Almost every argument about crypto wallets is really this argument wearing a costume. Not which chain, not which interface, just: when a transaction is signed, whose key signs it? Everything else, including every security property you care about, follows from the answer.

Custodial vs non-custodial wallets: the actual distinction

A blockchain has no accounts in the way a bank does. It has addresses, and the ability to move value from an address is granted by a private key. Whoever holds that key controls the funds. Full stop.

So “custody” reduces to one question: who holds the key?

In a custodial wallet, the provider does. Your balance in their app is an entry in their internal database saying they owe you. When you “send” crypto to another user of the same service, often nothing touches the blockchain at all, they update two rows. Your money is a claim, and its safety is the safety of the company.

In a non-custodial wallet, the key is generated on your device and stays under your control. The app is an interface to the blockchain, not an intermediary. The provider cannot move your funds. It also cannot stop anyone else who obtains your key.

How to tell which one you’re actually using

Marketing language is unreliable here. “Secure”, “your wallet”, “you’re in control” appear on both kinds of product. Ignore all of it and ask one question:

If I lose access completely, can the provider give me my funds back?

If yes, they can move your money without your key, which means they can also do it under a court order, a compliance flag, or an attacker who compromises their systems. That’s custodial. If the honest answer is “we cannot, because we never had the ability”, it’s non-custodial.

We’ve written the long version of that test, including what each answer implies and the cases where it’s genuinely ambiguous, in what a non-custodial wallet actually is. The rest of this page assumes you can already tell them apart, and gets on with the more useful question: which one you should be using.

How custodial and non-custodial wallets each fail

Both models work fine until they don’t. What matters is the shape of the failure.

Custodial failure

Your keys are fine; the company isn’t.

• Insolvency. Several large custodians have failed with customer funds commingled or missing. In bankruptcy, “your” crypto may be treated as an unsecured claim against the estate, meaning you join a creditor queue for an asset you believed you owned.
• Compromise. A custodian is a concentrated target holding many users’ keys. Exchange breaches have repeatedly produced nine-figure losses.
• Freezes and restrictions. Withdrawals get suspended for compliance review, during market stress, or because the provider has a liquidity problem. Your funds are visible and immovable.
• Jurisdiction. Access can change because of where you live, not anything you did.

Non-custodial failure

The company is irrelevant; you’re the single point of failure.

• Lost access. Traditionally a lost seed phrase meant permanently lost funds. No recourse, no appeal.
• Phishing and social engineering. Nobody validates the transaction you’re about to sign. A convincing fake site can obtain a signature that drains an address in one action.
• Irreversibility. A transfer to the wrong address is final. There is no reversal mechanism because there is no one operating a reversal mechanism.
• Malware. A compromised device can compromise a key.

The honest case for custody

Self-custody advocacy tends to skip this, and it shouldn’t.

Custodians provide real value. They absorb the consequences of user error, the wrong-address transfer, the forgotten password, the compromised phone. They provide fraud monitoring, human support, and in regulated markets sometimes insurance or segregation requirements. For someone buying a small amount as an experiment, a custodial account is very likely the correct choice, and the friction of key management would deliver no benefit proportional to the effort.

The case against custody is not that custodians are bad. It’s that custody quietly reintroduces the exact dependency that made a permissionless settlement network interesting in the first place. If your holdings can be frozen at a third party’s discretion, you have a slower, more volatile bank account.

Why the choice used to be so painful

For most of crypto’s history, choosing self-custody meant accepting seed phrases: twelve words whose loss was fatal and whose disclosure was fatal, to be kept perfectly and forever.

That is a genuinely bad deal for most people, and it explains the market outcome. Given “a company might fail” versus “you might forget”, most people sensibly picked the failure mode with a support line. The industry then spent a decade characterising this as user apathy rather than a rational response to a bad design.

Seed phrase vs social recovery

Why the twelve-word backup became standard, what it actually protects against, and what replaces it.

→

The choice is no longer binary

The useful development of the last few years is that “who holds the key” and “can I recover access” have come apart. They were only ever coupled by implementation, not by cryptography.

Social recovery lets you restore access through a quorum of guardians - your most used apps, your own second device, an identity provider you already use. No single guardian can move your funds. Enough of them together can help you regain control.

Multi-party computation splits key material so that a complete key never exists anywhere. Shares cooperate to produce a signature. There is nothing to steal in one place, and nothing to lose in one place.

Both keep the defining property intact, the provider still cannot unilaterally move your funds, so it is still non-custodial while removing the requirement that you personally safeguard one secret forever.

Which should you choose?

Rather than a ladder of amounts, the useful framing is a single comparison: which failure are you more exposed to?

You are more exposed to custodial failure if you hold meaningful value with one provider, if that provider is offshore or lightly regulated, if you’d struggle to absorb a months-long withdrawal freeze, or if your jurisdiction is one where accounts get restricted for reasons unrelated to your conduct. These are risks you cannot mitigate by being careful, because they aren’t about you.

You are more exposed to non-custodial failure if you have lost a phone in the last few years, if you’d store a backup in a photo roll or a notes app, if nobody else knows how to reach your assets if something happens to you, or if you routinely click through prompts without reading them. These risks are about you which means, unlike the first list, you can actually design around them.

That’s the real decision, and it’s why the modern answer is usually neither pure option. A wallet where you hold the keys but losing one thing is survivable removes the second list without reintroducing the first. Sizing matters too: keep spending money somewhere convenient and savings somewhere hard to reach, so that a worst case in either direction is annoying rather than catastrophic.

The question worth asking about any wallet isn’t whether it’s “secure”. It’s what happens when something goes wrong, and who gets to decide. If the answer is “the provider decides”, you should at least know that’s the deal you took.

Sizing it properly

How much to keep in a hot wallet, when cold storage starts earning its friction, and why most people should run both.

→