The actual difference
An exchange account is a record in a company's database saying you are owed a certain amount. The company holds the assets. Because a company is in the middle, there is someone to appeal to: a forgotten password can be reset, a stolen login can be frozen, a mistaken transfer can sometimes be traced and occasionally recovered. There is also someone who can freeze the account, close it, or fail.
A wallet is a key. Whoever has the key controls the funds, and the network does not know or care who that is. There is no password to reset, because there is no account — there is a cryptographic key and everything derived from it. Nobody can freeze your funds, and nobody can help you when something goes wrong.
Both of those descriptions contain a promise and a threat, and they are the same sentence read twice. Self-custody removes the intermediary who can fail you and, in the same motion, removes the intermediary who can save you.
| Exchange account | Self-custody wallet | |
|---|---|---|
| Who holds the asset | The company | Whoever holds the key |
| Lost credentials | Recoverable through identity verification | Permanent loss |
| Sent to the wrong place | Sometimes recoverable, slowly | Gone unless the recipient returns it |
| Platform fails or freezes you | Your problem | Not applicable |
| Compromised by malware | Second factor may save you | Usually total and immediate |
| Cost of a routine transfer | Internal transfers often free | Network fee every time |
Why both screens look the same and mean different things
A wallet app and an exchange app both show a list of assets and a number next to each. That visual similarity does a lot of quiet damage, because it suggests the two are variants of the same product.
They are not. The exchange balance is a claim, and moving it internally is a database update — instant, free, reversible by the operator. The wallet balance is a reading of the chain, and moving it is a transaction that costs a fee, takes time and cannot be undone by anyone.
The number that matters most on a wallet screen is often the one that is not displayed prominently: the balance of the chain's native coin, which is what pays for every action. A wallet holding a large stablecoin balance and no native coin is a wallet that can receive and cannot send. That is one of the most common ways people find themselves temporarily stuck, and it surprises them precisely because an exchange never behaves that way.
The seed phrase is the whole thing
What it actually is
The twelve or twenty-four words are not a password to the wallet. They are the mathematical origin of every key and address the wallet has and ever will have. Reinstalling the app and typing them in anywhere in the world reproduces the same wallet exactly, which is why they are a backup — and why they are also a complete theft if anyone else has them.
Anyone with the phrase has the funds. Not access to a copy of the wallet: the funds. There is no second factor behind it, no device binding, no alert. This is a different security model from anything else most people use, and treating it like a password is the root of most losses.
How people actually lose it
- Photographing it. The photo syncs to a cloud album, and the cloud account has a password that has appeared in a breach.
- Typing it into a "wallet validation" page after a support conversation that began with someone contacting them.
- Storing it in a notes app or password manager that is itself protected by a reusable password.
- Writing it down once and losing the paper. The quiet one. No attacker, no mistake, just a house move and a box that did not arrive.
- Restoring from a backup that was one word wrong. Discovered years later, when it is the only copy.
How to store it
Offline, on something durable, in more than one physical place, and never in anything that syncs. Two paper copies in two locations beats one perfect copy in one location, because the realistic threat for most people is loss rather than a burglar who knows what they are looking at.
Verify the backup once, immediately, by restoring it into a fresh wallet instance while the funds are still small. A backup you have never tested is a hypothesis.
The kinds of wallet, and what changes between them
- Browser extension. Convenient for interacting with on-chain applications, and living in the most attacked piece of software on your computer. Fine for small working balances; a poor place for savings.
- Mobile app. Better isolation than a browser extension on most phones, and easier to lose along with the phone if the seed phrase was never written down.
- Hardware wallet. The key never leaves the device, and transactions are confirmed on a screen the computer cannot alter. This closes off the entire category of malware that steals keys from a machine.
- Exchange-branded "Web3 wallet" products. These sit inside an exchange app and are usually genuinely self-custodial, with the keys split or held on your device rather than by the company. Read what the product actually says about who can recover the keys, because the answer determines which half of this article applies to it.
- Multi-signature and split-key setups. Require more than one approval to move funds. Meaningfully safer against both theft and single-point loss, and meaningfully more complicated to operate — the failure mode is losing enough of the pieces to be locked out.
The choice is less about which is "best" and more about matching the tool to how much is at stake and how often you need to touch it.
Token approvals: the risk nobody expects
Interacting with an on-chain application usually requires granting that application permission to move a token on your behalf. Many interfaces request an unlimited allowance by default, because it saves the user a transaction later.
That permission does not expire. It stays live after you have finished using the application and forgotten about it. If the application is later compromised — or was malicious from the start and simply patient — the permission is still there, and the funds can be moved without you signing anything new.
This is why wallets can be drained months after the interaction that caused it, with no recent activity on the user's part to explain it. Two defences: grant the specific amount rather than an unlimited one where the interface allows it, and periodically review and revoke the approvals attached to your address. Every major chain has a public tool for doing so.
A third defence is structural: use a separate wallet for interacting with applications, and never grant approvals from the wallet holding your savings. An approval can only reach what the granting address holds.
The attacks that target wallets specifically
- The fake support representative. Appears in a reply to a public question about a wallet problem, always contacts you first, and the conversation ends at a page asking for the seed phrase. No legitimate support ever needs it, because it is not an access credential — it is the funds.
- Address poisoning. A tiny transaction is sent to you from an address crafted to resemble one you use, so that it appears in your history. Later you copy the address from your own transaction list, and it is theirs. Copy from your address book, never from history.
- Blind signing. A transaction request that does not clearly say what it does. If the wallet cannot explain the action in plain language, rejecting it costs nothing and approving it can cost everything.
- The airdrop that requires a signature. Free tokens appear in your wallet, and claiming them requires connecting and signing. The signature is the point; the tokens are the bait.
- Fake wallet applications. Downloaded from an advertisement rather than the official site, and generating a seed phrase the author already knows. Install from the project's own domain, and verify the domain independently.
When a hardware wallet is worth it
The test is not a fixed amount, it is a question: if this balance disappeared tonight, would it change my life?
If yes, the cost of a hardware wallet is small against what it protects, and the protection is real rather than marketing — the key never touches the internet-connected machine, so the entire class of key-stealing malware is off the table.
What it does not protect against is you approving a malicious transaction. The device shows you what you are signing; if you confirm it anyway, it signs. Hardware moves the risk from "malware steals my key" to "I approve something I did not understand", which is a better risk but not no risk.
Buy from the manufacturer directly. A device that arrives with a seed phrase already written on a card is not a device, it is a trap, and this is a real and recurring scam on marketplace listings.
How to split between the two
The useful division is by function rather than by principle:
- On the exchange: money that is actively moving — trading, converting, on its way in or out. The recoverability is worth more than the custody risk for funds you are touching weekly.
- In self-custody: money that is sitting still and that you would be seriously hurt to lose. Nothing about a long-term holding benefits from being on a platform.
- In a separate application wallet: whatever you are actively using on-chain, funded per session, never holding more than you would accept losing to a bad approval.
The reason to split at all is that the two failure modes are uncorrelated. A platform incident does not touch your wallet; a compromised device does not touch the platform, provided the exchange account has a second factor that is not on that device. Concentrating in either one is the only way to lose everything at once.
Getting funds from one to the other is its own small subject — see which chain to withdraw on, because the transfer is where most of the avoidable losses in this article actually happen.
The problem nobody plans for
Self-custody is a single point of failure that includes you. If something happens to you, funds behind a seed phrase nobody else can find are gone permanently — there is no probate process that can reach them.
This is uncomfortable to plan for and worth doing anyway, at least to the level of ensuring that one trusted person knows a backup exists and where the instructions to find it are, without that being enough on its own to move the funds. Split-key and multi-signature arrangements exist partly for this reason. The exchange side has the opposite property: an account with verified identity has an inheritance process, however slow.
Questions people ask afterwards
Is a wallet safer than an exchange?
It removes one risk and adds another. Platform failure becomes impossible; your own mistakes become unrecoverable. Which is safer depends on which of those you are better equipped to prevent.
Can I use the same seed phrase on several chains?
Yes — one phrase typically derives addresses on many networks, which is why the same wallet shows balances on several chains. It also means one leaked phrase exposes all of them at once.
What if I lose my hardware wallet?
The funds are safe as long as the device had a PIN and you have the seed phrase. Restore into a new device and continue. The device is a container; the phrase is the wallet.
Is an exchange's own Web3 wallet really self-custody?
Usually yes in the technical sense, and the detail that matters is who can recover it. Read the product's own description of key recovery. If the company can restore your access without you, it is closer to a custodial product wearing self-custody language.
Can a transaction be reversed if I sign something malicious?
No. A signed and confirmed transaction is final. If an approval was granted rather than funds moved, revoking it immediately can prevent further loss, but anything already taken is gone.