In February 2025, hackers stole roughly $1.5 billion from Bybit, a single incident responsible for 44% of the $3.4 billion stolen from crypto platforms that year, according to ++Chainalysis's own 2026 Crypto Crime Report++. Bybit is an exchange, and it's diving into which kind of failure actually happened there, because it isn't the only kind a business needs to guard against.
"Custodial vs non-custodial" usually gets treated as a two-way choice: you hold your own keys, or someone else does. For a business deciding where to actually put treasury funds, there are four distinct arrangements worth knowing, and conflating the wrong two of them is a mistake that costs institutions real money.
Four models to know
- Self-custody means the business holds its own private keys directly, through hardware or software wallets it controls entirely. No third party can freeze, lose, or mishandle the funds, because no third party ever holds them.
- Exchange custody means funds sit with a trading platform as part of using that platform, typically in pooled, non-segregated wallets alongside other customers' balances. This is the category Bybit and Mt. Gox both fall into.
- Shared-control custody splits the key itself: the custodian holds part of it, the client holds another part, and neither side can move funds unilaterally. This is a distinct model from full third-party custody, controlled access with defined processes rather than either full delegation or full retention.
- Qualified third-party custody means a regulated, licensed custodian holds the assets on the business's behalf, with segregated accounts, independent audits, and typically insurance that covers the specific holding. Worth a precision note: "qualified custodian" has a specific defined meaning under US securities law, referring here descriptively to a regulated, audited custody provider generally, not a claim about meeting that specific US standard in every case.
MPC vs multisig: the two ways shared control actually works
Shared-control custody usually runs on one of two technical approaches, and they behave quite differently.
- Multisig requires multiple complete, distinct private keys, typically structured as a threshold like 2-of-3. Each signature is recorded on-chain, so the approval structure, the threshold, and which keys signed are all visible to anyone looking at the blockchain. Changing signers or the threshold generally requires an on-chain update, sometimes a new wallet address entirely, which makes it rigid but gives it a native, verifiable audit trail.
- MPC (multi-party computation) splits a single key into cryptographic shares distributed across parties, who jointly compute a signature without ever reassembling the full key. The resulting on-chain transaction looks like a standard single-signature transfer, so the number of participants and the approval policy stay private. It's more flexible, shares can be rotated without changing the wallet address, but the audit trail lives in off-chain logs rather than on the chain itself.
Multisig suits a business that wants on-chain, third-party-verifiable proof of who approved what. MPC suits a business that wants operational flexibility and doesn't need that approval history to be publicly visible.Whether an MPC setup counts as custodial or non-custodial depends entirely on who holds enough key shares to meet the signing threshold. If the business controls that threshold alone, it functions as self-custody even though a provider is technically involved.
Why exchange custody isn't one risk, it's two
Exchange custody failures don't all look the same, and the difference matters for what actually protects against each. Bybit and Mt. Gox were external hacks, an outside attacker breached custody infrastructure the exchange itself was trying to defend. FTX was different: Sam Bankman-Fried was convicted of fraud for misappropriating customer funds through Alameda Research, the exchange's own operator misusing assets it was supposed to be safeguarding.
Segregation and independent audits are the defense against the FTX pattern, in theory, though FTX's own auditors failed to catch the problem before collapse, which is itself worth knowing. Cold storage, withdrawal limits, and MPC-based key splitting are the defense against the Bybit pattern.
A business evaluating exchange custody risk needs to ask about both, not assume one set of controls covers both threats.
Where safekeeping fits as its own regulated category
Custody isn't a side feature of a crypto business, it's a distinct activity that regulators license separately. ++FATF's guidance on virtual asset service providers++ names "safekeeping and administration" as one of five recognized VASP activity categories, separate from exchange and transfer services. A provider offering trading, custody, and settlement under one roof may hold, or need to hold, multiple distinct authorizations to legally offer all three.
For a business evaluating a provider, this is a concrete question worth asking directly: is custody something this provider is specifically licensed for, or is it a byproduct of them also running an exchange?
What custody insurance actually covers, and doesn't
Custody insurance policies typically respond to a fairly narrow set of events: theft of private keys through external hacking, insider theft or collusion by a custodian's own employees, fraudulent transfer of assets from custodian wallets, and physical destruction of cold-storage hardware.
What most policies exclude: market volatility, protocol or smart-contract exploits, war and government seizure, and, critically, key mismanagement by the asset owner. That last exclusion matters most for self-custody specifically. Many standalone self-custody insurance policies explicitly exclude losses caused by the policyholder's own key mismanagement, which is exactly the risk self-custody carries the most.
Coverage structure also scales with holdings. Large portfolios typically carry aggregate limits with per-incident sublimits underneath them, so the headline coverage number doesn't necessarily mean any single loss event is covered in full. Sophisticated treasury teams often layer coverage rather than relying on one policy: custodian-level crime insurance for assets under qualified custody, a separate specie policy for anything self-custodied, and entity-level D&O or E&O coverage if the holdings sit inside a fund or operating company. No single policy on the market today covers the full range of realistic loss scenarios.
Exchange custody sits in the weakest position of all four models here. Pooled, non-segregated wallets make it difficult for any insurer to write a policy that protects an individual customer's specific holdings, since there's no clean legal separation between one customer's funds and another's.
Decision framework
| Factor | Favors self-custody | Favors shared-control | Favors qualified third-party custody |
| Balance size | Smaller, operationally manageable | Mid-size, active treasury use | Larger, where loss would be catastrophic |
| Operational capability | In-house technical and security expertise | Some internal capability, wants a check on unilateral access | No dedicated blockchain security team |
| Liquidity needs | Frequent, active movement of funds | Frequent movement, wants dual control | Treasury holdings, less frequent access |
| Regulatory obligations | Fewer external compliance requirements | Moderate, some audit trail needed | Fiduciary duty to clients or investors, audit requirements |
| Insurance needs | Coverage is narrow and often excludes the main risk | Split-key structure can simplify underwriting | Needs verifiable, third-party-underwritten coverage |
Exchange custody doesn't appear as a favorable column here deliberately. It's rarely the right choice for treasury funds specifically, since it combines the operational convenience of a third party holding the assets with meaningful counterparty risk, without the compliance depth of a provider actually licensed for safekeeping as its core business.
Related reading: Digital Asset Custody · What is crypto custody? · What is institutional crypto custody?