Self-Custody: The Real Tradeoffs Between Wallet Types

Self-custody discussions tend to collapse into a ranking — hardware wallet good, exchange bad, multisig best. That ranking is misleading, because the options do not differ in how secure they are so much as in what they are secure against. A setup that eliminates one failure mode routinely introduces another.

Start with the failure modes, not the products.

The four ways people actually lose coins

Remote compromise. Malware, a malicious browser extension, a poisoned download, or a signature obtained under false pretences. The attacker never touches you physically.

Physical theft or coercion. Someone gets hold of your device, your written seed phrase, or you. Sometimes called the “$5 wrench attack” — the observation that cryptographic strength is irrelevant if the holder can simply be threatened.

Loss of access. No adversary at all. A seed phrase destroyed in a flood, written down with a transcription error, stored somewhere the holder later cannot recall, or protected by a passphrase only the holder knew — and who has since died. This is, by a wide margin, the most common cause of permanent loss.

Operator error. Sending to the wrong address, signing a transaction that did something other than advertised, or approving an unlimited token allowance to a contract that later drains it.

Notice that only the first is a “hack.” Most self-custody advice optimises hard against remote compromise and quietly makes loss of access more likely.

How the seed phrase works

Nearly all modern wallets derive every key they will ever use from a single master secret, displayed to you as twelve or twenty-four words. The word list and encoding are standardised, which is why a seed can generally be restored into a different vendor’s wallet — a genuinely important property, because it means your recovery does not depend on a company continuing to exist.

Two details are routinely misunderstood.

The words are not a password protecting the keys. They are the keys, in a transcribable form. Anyone who reads them has your funds, immediately and irreversibly.

And the derivation path matters. A restored seed produces the correct addresses only if the restoring wallet uses the same derivation scheme. Restoring into a wallet that defaults to a different path can show an empty balance while the funds are perfectly safe at addresses the wallet is not looking at — a recoverable situation that has caused a great deal of unnecessary panic.

Hot wallets

A wallet whose keys live on an internet-connected device: a phone app, a browser extension, a desktop client.

Strong on convenience and on operator visibility — you can read what you are signing on a full-sized screen. Weak against remote compromise, because the signing key exists in the memory of a machine that also runs everything else you do.

Reasonable for amounts you would be annoyed but not damaged to lose. The mistake is not using one; it is using one for a life-changing balance.

Hardware wallets

A dedicated device that holds the key and signs transactions internally, so the key never reaches the connected computer.

This is a targeted defence against exactly one thing: remote compromise. It is highly effective at that. It does nothing about physical theft — most devices are protected only by a PIN — and nothing about coercion.

Its most underrated feature is the screen. Verifying the destination address on the device itself defeats clipboard-hijacking malware, which silently swaps a copied address for the attacker’s. A hardware wallet whose owner blind-approves whatever appears has given up much of the protection it was bought for.

The optional passphrase (sometimes called the 25th word) is a genuine security upgrade and a genuine loss-of-access hazard. It creates an entirely separate wallet from the same seed. Forget it and the funds are gone, with no recovery path — the seed phrase alone will not reach them.

Multisig

A quorum arrangement: funds move only when m of n keys sign — 2-of-3 being the common choice.

This is the only configuration that meaningfully addresses physical theft and coercion. A thief who takes one device gets nothing. It also removes single points of failure in backup: lose one key of three and you still recover.

The cost is complexity, and complexity is not a footnote here. You now maintain multiple keys in genuinely separate locations, plus the wallet descriptor recording the quorum — which is required for recovery and which people routinely forget to back up. On-chain fees are higher, because more signatures mean a larger transaction. And your heirs must understand a scheme most people have never encountered.

Multisig converts a security problem into an operations problem. For some holders that is an excellent trade. For someone who will not maintain it, it raises the risk of loss of access above the risk it removed.

Backups are the part that gets skipped

Paper burns and runs. Metal backup plates exist because of this, and are cheap relative to what they protect.

More importantly: an untested backup is not a backup. The only way to know your recovery works is to perform it — restore the seed onto a spare device, confirm the expected addresses appear, then wipe it. People discover transcription errors during a real emergency remarkably often, and an emergency is the worst possible time to discover one.

Plan for inheritance explicitly. Perfect operational security that dies with you is indistinguishable, to your family, from having lost the funds.

Choosing

• Small, frequently spent balance → hot wallet, accept the risk knowingly.
• Meaningful balance, single holder → hardware wallet, verify on-device, metal backup, tested restore. A passphrase only if you have a durable plan for remembering it.
• Large balance, or a real physical-threat model, or shared control → multisig, but only if you will genuinely maintain it.

The best setup is not the most sophisticated one. It is the most sophisticated one you will still be operating correctly in five years.

Educational content only. Nothing here is financial advice.