Crypto Custody in 2026: Control Is Not the Same as Ownership

Two events bookend crypto custody in 2026, and they point in opposite directions. In January, BitGo listed on the New York Stock Exchange and became the first public, federally chartered digital-asset infrastructure company, closing its initial offering at $18.00 per share alongside final approval to operate as a national trust bank. In July, the FTX estate began its fifth creditor distribution, roughly $900 million, pushing cumulative repayments toward $10 billion since payouts started in early 2025. One is a story about custody becoming a regulated, investable business. The other is a story about what happens when custody goes wrong. Between them sits the question almost every crypto holder answers too fast: who holds my coins?

The honest answer is that who holds the keys is only half the question, and usually the less important half. There are two separate things you can ask about any pile of crypto. First, who can technically move it right now? Second, if the party holding it disappears tomorrow, insolvent, hacked, or indicted, what do you actually own? Most comparisons, and most buyers, fixate on the first axis and ignore the second. Celsius and FTX customers learned, expensively, that the second axis is the one that pays out. This guide compares the full range of custody models on both axes. For price context, Bitcoin traded near $64,000 and Ether near $1,910 on 12 August 2026. Nothing here is investment advice, and the word best is meaningless until you say which failure you are trying to survive.

Two Axes, Not One

Start with a framework, because the models only make sense against it. Axis one is technical control: who possesses the signing capability that actually authorizes a transaction on-chain. Axis two is the legal claim: if the holder fails, are the assets segregated and yours (bankruptcy remote), or are you standing in line as an unsecured creditor holding an IOU? These two axes are independent. You can have full control and a rock-solid claim (you hold your own keys), or no control and a strong claim (a properly segregated custodian signs for you but the assets are legally yours), or no control and a weak claim (an exchange balance), or, rarely, control over assets that are contractually someone else’s.

The trap is that convenience clusters in the worst corner. The easiest experience in all of crypto, tapping deposit on an exchange app and watching a number appear, lands you in the low-control, weak-claim quadrant. It feels like a bank account. It is legally closer to a store credit. The table below maps each model onto the two axes; the rest of this piece walks through them one by one and then stress-tests them against the two bankruptcies that rewrote the rules.

Self-Custody: Total Control, Total Responsibility

Self-custody is the purest position on both axes: you hold the private keys, so you have complete technical control and the strongest possible legal claim, because there is no counterparty who could go bankrupt with your coins. This is the position the community shorthand captures in the maxim not your keys, not your coins, popularized by the late educator Andreas Antonopoulos. It is also the position with the fewest safety nets. Nobody can freeze your funds, and nobody can restore them either.

Self-custody splits into hot and cold. A hot wallet is software connected to the internet, convenient for daily use and exposed to phishing, malware, and malicious transaction approvals; the practical differences between the popular browser and mobile options are covered in our MetaMask, Phantom and Rabby comparison. A cold wallet keeps the key on a device that never touches the internet, which shrinks the remote attack surface but shifts the risk to you: the seed phrase. Lose it and the coins are gone; let someone photograph it and the coins are theirs.

The failure modes of self-custody are not exotic. They are human.

• Key loss: a forgotten passphrase, a discarded backup, a house fire. No recovery exists.
• Blind signing: approving a transaction whose real effect is hidden behind opaque calldata, the pattern behind several of the largest 2025 losses.
• Malware: clipboard hijackers that swap a pasted address, and seed-stealing trojans.
• Physical coercion: the rise of so-called wrench attacks, where a known holder is targeted in person, a threat no cold wallet defends against on its own.
• Death and incapacity: the single most neglected risk. If only you can access the keys, your heirs inherit nothing but a locked box.

Self-custody is the correct answer for the sovereignty-minded and for anyone who fears counterparty failure above all else. It is the wrong answer if the failure you should actually fear is your own mistake or your own mortality. That gap is exactly what the next two models try to close.

Collaborative Custody and Multisig: Splitting the Single Point of Failure

A single key is a single point of failure in both directions: one leak loses everything, one loss loses everything. Multisig fixes this by requiring several keys to authorize a transaction, a two-of-three or three-of-five arrangement enforced on-chain by a script or smart contract. No single key can move funds, so a thief needs to compromise a quorum, and you can lose one key without losing access.

Collaborative custody productizes this for people who do not want to run three keys alone. In a typical two-of-three setup with a provider such as Unchained, Casa, or Nunchuk, you hold two keys and a key agent holds one. The agent cannot move your funds, because one signature is not a quorum; it exists to co-sign when you request it and to help you recover if you lose a key. Some providers add structured inheritance and identity-based recovery on top, turning the death-and-incapacity problem from a hope into a documented process. On the two axes, you keep effective control and the strongest legal claim, while outsourcing only the operational fragility.

The trade-off is complexity and configuration risk. Multisig is transparent and verifiable on-chain, but it is chain-specific, it costs more in fees, and a misconfigured signing setup or a compromised signing interface can still fail. Multisig removes the single point of failure; it does not remove the need to understand what you built.

MPC: One Signature, No Single Key

Multi-party computation, or MPC, reaches a similar goal by different math. Instead of several distinct keys signing on-chain, MPC splits a single key into shards that are distributed among parties and never assembled in one place. The parties jointly compute a signature off-chain, and the network sees just one ordinary signature. Because the full key is never reconstructed, there is no single secret to steal, and because the on-chain footprint is a normal signature, MPC is chain-agnostic and works the same across dozens of networks, with programmable policy controls layered on top (spending limits, approval workflows, allowlists).

This is why most institutional custodians and many exchanges run on MPC rather than script-based multisig: it is more flexible, cheaper to operate across chains, and easier to wrap in policy. The cost is transparency and trust surface. Multisig is verifiable on the public ledger; MPC happens inside a provider’s software, so you are trusting an implementation you cannot fully inspect, and threshold-signature libraries have had their own vulnerabilities. MPC and multisig are not a ranking. They are different tools with different transparency and portability profiles, and the strongest 2026 setups often combine them, using MPC for operational signing and multisig or hardware for the deepest reserves.

Exchange Custody: The Omnibus Account and the IOU

When you deposit crypto on a centralized exchange, you do not get a personal vault. Your coins are pooled into commingled omnibus wallets the exchange controls, and your balance becomes a row in the exchange’s database, a claim on the firm rather than a specific set of coins. You have surrendered both axes at once: the exchange can move the assets, and what you hold is an IOU. In return you get genuine convenience, instant trading, password recovery, no seed phrase to lose, and a compliance layer that handles the identity and anti-money-laundering checks that self-custody pushes onto you.

For an active trader keeping a working float, that trade is often reasonable. The danger is treating an exchange balance as savings. In the low-control, weak-claim quadrant, the thing that determines your fate is not the exchange’s marketing but the small print of the account you opened and the solvency of the firm. That is abstract until a bankruptcy makes it concrete, which is what the stress test below is for. Note the distinction that trips people up: an interest-bearing or yield account is not custody at all. It is a loan you made to the platform, a point we map out in detail in our look at whether on-chain credit is safe. Custody keeps your assets aside; lending puts them to work and puts you at the back of the line.

Qualified and Institutional Custodians: The Regulated Tier

Above the retail exchange sits a tier built specifically to fix the legal axis. A qualified custodian, in United States terms, is an entity that meets the standard under the Investment Advisers Act custody rule, so that registered funds and advisers can hold client crypto with it. The defining feature is not fancier storage; it is legal structure. These custodians are typically chartered as trust companies or national trust banks, which means client assets are held in a fiduciary capacity and segregated from the custodian’s own balance sheet. You have zero technical control, yet you sit in the strong-claim quadrant, because the assets are structured to be bankruptcy remote. That structure is the entire product.

Coinbase Custody Trust Company is a New York limited-purpose trust supervised by the New York Department of Financial Services, acting as a fiduciary and a qualified custodian, with assets in cold storage and third-party SOC audits. Anchorage Digital broke the trail in 2021 as the first federally chartered crypto bank approved by the Office of the Comptroller of the Currency, able to operate across all fifty states. BitGo joined at the national-charter level and then went public in January 2026. Fidelity Digital Assets, a New York trust company, custodies for Fidelity’s own funds. The table below compares the tier on the dimension that matters, regulatory footing, rather than on marketing.

Bank Custody: TradFi Arrives at the Vault

The biggest structural change of 2026 is that traditional banks are now real custody competitors, not press-release ones. The clearest signal is Standard Chartered moving to fully absorb Zodia Custody, the digital-asset custodian it co-founded, folding regulated custody directly into the bank’s securities-services arm, with completion targeted for the end of August. Zodia chief executive Julian Sawyer framed the shift bluntly to CoinDesk, arguing that “every single bank will soon need to hold digital assets.” Deutsche Bank is building custody through partnerships, and other large banks have moved from studying the question to answering it.

Two regulatory changes made this economic rather than aspirational. First, the Securities and Exchange Commission rescinded Staff Accounting Bulletin 121 in January 2025 and replaced it with SAB 122. The old rule effectively forced a custodian to book safeguarded client crypto as a liability on its own balance sheet, a capital penalty that made bank custody uneconomic; removing it reopened the door. Second, the Office of the Comptroller of the Currency confirmed through interpretive letters in 2025 that national banks may provide crypto custody, and later that they may buy and sell custodied assets and outsource execution, without first seeking a supervisory non-objection. On the two axes, bank custody with proper trust structuring can land in the strong-claim quadrant. But note the sharp limit that catches many newcomers: crypto held at a bank is not a bank deposit, so it is not covered by federal deposit insurance, a point the next section makes precise.

The ETP Wrapper: Custody You Never Touch

The fastest-growing custody model is one where you never hold crypto at all. A spot Bitcoin or Ether exchange-traded product turns the asset into a security: you buy shares in a fund through an ordinary brokerage account, and the fund hires a custodian, very often Coinbase, to hold the underlying coins. On the control axis you are at zero, and on the asset axis you do not even hold the crypto; you hold a regulated security whose value tracks it. That is not a flaw so much as a different product for a different buyer.

The wrapper suits investors who want price exposure inside a retirement account or a brokerage statement, with familiar tax reporting and the investor protections that attach to registered securities. It is a poor fit for anyone who wants to actually use the asset, because you cannot withdraw coins, move them on-chain, self-custody them, or spend them. You are also stacking counterparty layers: the fund sponsor, the custodian it selects, and the market maker. The ETP does not remove custody risk. It relocates it into a securities wrapper and hands the operational problem to a professional, which for a large share of buyers is precisely the point.

The Insolvency Stress Test: What Celsius and FTX Actually Taught

Here is where the second axis earns its keep. The best way to compare custody models is not to ask which is most secure on a good day, but which pays out on the worst day. Two bankruptcies supply the case law.

Celsius is the cleaner lesson. When the lender collapsed, United States Bankruptcy Judge Martin Glenn ruled in early 2023 that assets in the platform’s Earn accounts were property of the bankruptcy estate, based on the plain language of the terms of use, which meant those customers became unsecured creditors of a bust company. Holders of the platform’s separate Custody accounts were treated as owners of their assets and had them returned. Same company, same app, two account types, opposite outcomes. Not the technology and not the brand, but the contract you clicked through decided who owned the coins.

FTX teaches a subtler lesson about what recovery even means. The estate has now repaid an extraordinary amount, and by its fifth distribution many creditor classes are above 100% of their allowed claim in dollar terms, with the smallest convenience class reaching 120%. That sounds like customers were made whole. They were not, in the way that matters to a crypto holder. Claims were valued (dollarized) at the petition-date prices of November 2022, near the bottom of the cycle. Someone owed one Bitcoin got the November 2022 dollar value of that Bitcoin plus interest, not a Bitcoin, and missed the entire rally to today’s levels. A 105% recovery is a heavy loss if the asset tripled while your claim was frozen in dollars. The table below lines the precedents up.

The combined lesson is uncomfortable and simple: in the low-control quadrant, your recovery is governed by an account type and a valuation date you probably never read. Custody is a legal question wearing a technical costume.

Insurance, SIPC and FDIC: What Actually Pays

A lot of comfort in this market rests on a misunderstanding about insurance. Neither of the two familiar United States backstops applies to crypto. Cryptocurrency is not a bank deposit, so the Federal Deposit Insurance Corporation does not cover it, and it is generally not treated as a registered security, so the Securities Investor Protection Corporation does not protect it either. When a platform advertised itself as protected by these programs during the last cycle, that framing drew regulatory fire precisely because it was misleading. The insurance that does exist at reputable custodians is private, usually a specie or crime policy underwritten in the Lloyd’s market, and it has a crucial shape: it typically covers theft of assets from storage up to a stated sub-limit, not the full value under custody, and it does not cover market losses or the custodian’s own insolvency. Read what the policy insures against before you treat it as a guarantee, because the gap between marketing insured and actually insured is where customers get hurt.

Proof of Reserves: A Partial Answer

After FTX, proof of reserves became the industry’s favorite reassurance. The standard technique, a Merkle-tree attestation, was popularized by Castle Island Ventures partner Nic Carter after the Mt. Gox collapse, and it lets an exchange demonstrate that it controls enough on-chain assets to cover the customer balances included in the snapshot, without exposing individual accounts. It is genuinely useful, and it is genuinely incomplete. As Carter has argued at length, proving reserves is only half a balance sheet. It shows assets on one day; it does not show liabilities, so an exchange can display large holdings while hiding debts, off-balance-sheet obligations, or coins borrowed just for the snapshot. The basic scheme can even be gamed by constructing the liability tree dishonestly. A credible proof of reserves needs a matching proof of liabilities, ideally under an auditor, and the field is moving toward zero-knowledge techniques to close the gaps. Treat a proof-of-reserves badge as one input, not a clean bill of health.

The 2026 Regulatory Map: MiCA Segregation vs the US Patchwork

Where you custody increasingly decides how strong the second axis is, because the two big jurisdictions took different routes. The European Union wrote segregation into statute. Under Article 75 of MiCA, a licensed custody provider must keep client crypto legally and operationally segregated from its own estate, so that the provider’s creditors have no recourse to client assets in an insolvency, and it is liable for the loss of client assets or the means of access from an incident attributable to it, capped at market value at the time of loss. That is a default of bankruptcy remoteness by law, the exact protection Celsius Earn customers did not have.

The United States has no single federal custody statute for crypto. Instead it runs a patchwork: the Advisers Act qualified-custodian rule for funds, OCC national trust charters, New York trust licenses, and the Securities and Exchange Commission as the reference markets regulator, with the boundaries of liability still being fought out case by case, a landscape we map in our piece on who is actually on the hook in 2026. The practical upshot for a holder is stark. In the EU, segregation is largely the legal default for a compliant provider. In the US, whether your assets are bankruptcy remote depends on how the specific custodian is chartered and what your account agreement says, which is why the account type, not the logo, remains the thing to check.

How to Choose: Match the Model to the Failure You Fear

Stop asking which custody model is most secure, because that question has no answer. Ask which failure you are insuring against, because each model is strong against one failure and weak against another. Line up the fear with the model.

• Afraid an exchange or lender goes bust: do not keep more than a trading float on any platform, and hold savings in self-custody or with a segregated, properly chartered custodian.
• Afraid you will lose your own keys or die without a plan: use collaborative custody or a qualified custodian, and consider recovery-oriented setups such as the social recovery that smart-account wallets enable.
• Afraid of being physically targeted: favor self-custody with keys split across locations, and do not advertise your holdings.
• Afraid of your own operational mistakes: choose managed custody with policy controls, spending limits, and approval workflows.
• Managing an institution or a fund: use a qualified custodian or bank for the legal segregation, and lean on MPC for day-to-day operations.

A simple size heuristic sits underneath all of this. Small and active balances can reasonably live in a hot wallet or on an exchange. Meaningful savings belong in self-custody or collaborative custody. Large or fiduciary holdings belong with a regulated custodian whose segregation you have actually verified. And custody, like the assets themselves, rewards diversification: splitting holdings across a self-custody setup and a segregated custodian means no single failure, technical or legal, takes all of it. The models on this page are not a ladder from bad to good. They are a menu of trade-offs, and the only wrong choice is the one made without reading the second axis.

Frequently Asked Questions

What is the safest way to store crypto in 2026?

There is no single safest option; the right model depends on which failure you are trying to survive. Self-custody removes counterparty risk but puts key loss, theft, and inheritance entirely on you. A qualified custodian or a segregated CASP adds legal protection in a bankruptcy but reintroduces counterparty risk. For most people the practical answer is to keep only a trading float on an exchange and hold long-term savings in self-custody or collaborative custody.

Does FDIC or SIPC insurance protect my crypto on an exchange?

No. Cryptocurrency is not a bank deposit, so FDIC insurance does not apply, and it is generally not a registered security, so SIPC protection does not apply either. The only backstop is the firm’s own private insurance, which usually covers theft up to a sub-limit and does not cover the firm going bankrupt.

What did the Celsius bankruptcy teach about crypto custody?

That the account type, not the brand, decides ownership. A US bankruptcy judge ruled that assets in Celsius Earn accounts were property of the estate, so those users became unsecured creditors, while holders of dedicated Custody accounts were treated as owners and had assets returned. The terms of service determined who owned the coins.

Is MPC custody better than multisig?

They are different tools rather than one being strictly better. MPC splits a key into shards that are never assembled, signs off chain, produces a single on-chain signature, and works across many chains with policy controls. Multisig requires several on-chain signatures, is transparent and verifiable on chain, but is chain specific. Institutions often prefer MPC for flexibility; both are still software.

Are Bitcoin ETFs a form of custody?

Yes, an indirect one. You own shares in a fund, and the fund hires a custodian, often Coinbase, to hold the underlying coins. You have no technical control and cannot withdraw the crypto itself, but you get brokerage and tax integration plus the protections that apply to a regulated security.

By Marcus Okafor, senior markets editor at HOGE Wire, covering custody, market structure, and the plumbing of crypto finance.