Custody is a risk choice

The choice is between two lists of risks

Custody debates are usually conducted as ideology, which is unhelpful because both options carry real and different risks. The useful framing is not which is safer in the abstract but which set of failures you are better placed to prevent.

With provider custody, a service holds the keys and can restore your access. You are exposed to its security, its withdrawal controls, its solvency, its jurisdiction, its willingness to serve you, and its support process on a bad day. You are also exposed to your own account security, because a provider will generally honour a login that looks legitimate.

With self-custody, you hold the keys and remove the provider from the picture. You are now exposed to your own backups, your device security, your ability to read a transaction before approving it, your plan for what happens if you are unavailable, and the permanence of every mistake.

Which failures each model makes possible

Provider custody makes some catastrophes impossible: you cannot lose everything by misplacing a piece of paper, and a mistyped address can sometimes be recovered internally. It makes others possible: an account freeze during a dispute, a withdrawal halt during stress, or an insolvency in which your balance becomes a claim in a legal process.

Self-custody inverts this. No company can freeze you, and no company's failure touches your assets. In exchange, an unrecoverable backup, a wrong address, or one careless approval produces a loss with no appeal. There is no dispute process because there is no counterparty.

Notice that neither list is shorter. The realistic comparison is which failures you can actually reduce through effort, and that depends on the person rather than on the technology.

A staged approach

The common mistake is treating this as a single decision made once, at maximum stakes. Someone reads that self-custody is safer, moves everything at once, and turns their first attempt at key management into a test they cannot afford to fail.

A better sequence is gradual. Learn with an amount whose total loss would be annoying rather than serious. Practise a full recovery from the backup before it matters. Keep an activity wallet separate from long-term holdings so an unfamiliar application never touches the latter. Increase what you hold yourself only as each step is proven, and keep the option of a reputable provider for the portion you are not yet ready to secure.

Splitting across both models is not indecision. It is a reasonable response to the fact that the two carry uncorrelated risks.

• Name the specific failure you are most likely to cause yourself.
• Rehearse recovery before value depends on it.
• Keep learning funds separate from long-term funds.
• Enable phishing-resistant authentication on any provider account.
• Re-examine the split after any life or portfolio change.