In the first half of 2025, $2.17 billion was stolen from crypto. The largest single theft—$1.34 billion from Bybit—targeted an institution with sophisticated multi-signature cold storage. Here’s what actually went wrong, and a framework for getting custody right at institutional scale.
Table of Contents
The Moment Custody Became a Board-Level Question
February 2025. Bybit, one of the largest crypto exchanges in the world, lost $1.34 billion in a single attack, the largest crypto theft in history. Every headline called it a “multisig failure.” That framing is wrong. Understanding why it’s wrong is the starting point for every custody decision worth making today.
First, the scale of the broader problem. In H1 2025, Chainalysis tracked $2.17 billion stolen across the crypto industry. In all of 2024, private key compromises accounted for 43.8% of total losses. These weren’t smart contract bugs or protocol exploits. They were failures at the key management layer.
What shifted in 2025 is who the targets are. The victims are institutions with compliance teams, security budgets, and sophisticated wallet setups. The custody question has graduated from a personal finance decision to a board-level architectural one.
The regulatory and technical landscape caught up in parallel. The US clarified institutional custody standards in H2 2025 more than it had in the previous five years combined. New custody technology is maturing. And the lessons from 2025’s worst incidents provide a clearer framework than any theoretical analysis could.
By the end of this article, you’ll have a framework for which custody architecture fits your specific institutional context, whether you run a crypto-native fund, a corporate treasury making its first BTC allocation, or a company operating customer crypto products.
How to Get Free Crypto
Simple tricks to build a profitable portfolio at zero cost
The Custody Spectrum
Custody runs across a spectrum of five approaches. Most serious institutional setups combine more than one.
Fully custodial. A third-party service holds your private keys. Lowest operational friction, highest counterparty risk. FTX is the proof: when the custodian fails, the assets go with it.
Self-custody, single key. You control the private key directly. A hardware device or a non-custodial software wallet like Bitcoin.com wallet, which gives users full multi-chain key control with no third-party access, keeps you in complete control. Practical for individuals and smaller holdings. This approach hits an operational ceiling above a few million dollars: one person, one key, one catastrophic point of failure.
Multisig. Multiple independent private keys must sign to authorize a transaction. A 2-of-3 or 3-of-5 threshold distributes control across people or devices. The industry standard for on-chain shared treasury management and the approach at the center of the Bybit case.
MPC wallets. Multi-Party Computation (MPC) distributes cryptographic key shares across multiple parties. The full private key never exists as a complete object at any point. Chain-agnostic, lower gas costs, faster signing. Fireblocks, BitGo, and Cobo are the dominant institutional providers.
Third-party qualified custodian. A regulated institution—with a NYDFS trust charter, OCC federal charter, or equivalent—holds assets under fiduciary responsibility, with segregated accounts and insurance coverage. The closest analog to a traditional bank custody relationship.
These approaches combine in practice. The most common institutional architecture in 2025 uses MPC for operational assets, hardware cold storage for long-term reserves, and a qualified custodian for assets subject to LP or regulatory requirements.
The Bybit Lesson: What Actually Happened?
Every headline called it a multisig failure. The actual attack tells a more important story, and it changes the institutional implications entirely.
Bybit used a 3-of-3 multi-signature cold wallet managed through Safe (Gnosis Safe), the dominant smart contract wallet on Ethereum. Safe is battle-tested, open-source, and used by the majority of large DeFi protocols for treasury management.
The attacker—later attributed to Lazarus Group, North Korea’s state-sponsored hacking operation—didn’t break the multisig cryptography. They compromised Safe’s frontend infrastructure and served Bybit’s signers a spoofed transaction interface.
What the signers saw on screen was a routine internal transfer between wallets. What they were actually signing was a malicious transaction that replaced the Safe’s implementation contract with an attacker-controlled one, granting unlimited withdrawal access. Three legitimate signers, three valid signatures, and $1.34 billion gone. The cryptography worked exactly as designed but the human approval layer failed.
That distinction matters more than any technical comparison of multisig vs. MPC. The same attack—presenting legitimate signers with a fraudulent transaction interface—works identically against an MPC setup. If the participants in a 2-of-3 MPC threshold are presented with a spoofed transaction and approve it, the threshold signature executes it correctly. Key distribution is irrelevant here.
What actually prevents this class of attack: transaction simulation (before signing, you see exactly what the transaction does on-chain), policy enforcement (automated rules that block anomalous transactions before they reach signers), and UX-layer verification (confirming the signing interface matches expected parameters).
Those features live in the governance layer built on top of cryptographic key management. Consumer-grade wallets don’t have them. Institutional-grade custody platforms do, and the distinction that matters.
The broader lesson the Bybit case teaches: Security architecture has a perimeter problem. Most institutional custody investment goes into hardening the cryptographic interior—key generation, threshold requirements, hardware isolation. Far less goes into hardening the interface layer where humans actually interact with the system. As cryptographic models become harder to break directly, the interface layer becomes the natural attack surface. Institutional custody design needs to treat it accordingly.
Multisig: Genuine Strengths and Real Limits
Multisig gets mislabeled as the technology that failed at Bybit but it deserves a fairer assessment.
The genuine strengths are substantial. Multisig is on-chain and fully transparent—any auditor can verify the signature threshold without trusting a third party. Bitcoin’s P2SH (pay-to-script-hash) multi-signature format has operated without cryptographic failure since 2012. Gnosis Safe secures well over $100 billion in on-chain assets. Thirteen years of unbroken production track record is not a theoretical claim.
Multisig remains the right choice in specific contexts:
- on-chain DAO treasury management where public auditability is a governance requirement
- protocol admin key management where transparency to the community matters
- any setup where a chain-native, auditable signing structure outweighs the operational overhead.
Where it genuinely breaks down: chain specificity (a separate multisig setup is required for Bitcoin, Ethereum, Solana, and every other chain you operate on), gas costs at scale (each signature is an on-chain operation—expensive when compounded across thousands of transactions), signing coordination friction (getting three to five key holders synchronized for time-sensitive transactions is painful), and the UX-layer vulnerability the Bybit hack exposed.
The right conclusion isn’t “don’t use multisig.” Multisig is the right answer for specific use cases—and the wrong answer when treated as a complete custody solution without addressing the governance layer above it.
MPC: Why Institutions Are Moving Here, and What the Marketing Doesn’t Say
Multi-Party Computation has become the dominant institutional custody approach. It solves the problems multisig doesn’t—and introduces a few of its own.
The Core Architectural Advantage
The private key never exists as a complete object. Distributed Key Generation (DKG) creates key shares in isolation, across independent parties or environments. No single participant ever sees the full key during generation or signing. Threshold Signature Schemes (TSS) allow participants to collaboratively produce a valid cryptographic signature—and what reaches the blockchain is a standard single signature, indistinguishable from a traditional single-key wallet transaction.
The practical benefits follow from this architecture: chain-agnostic operation (one MPC setup works across every chain without separate configurations), lower gas costs (the blockchain sees a standard signature, not a multisig contract call), faster signing, and native support for complex policy engines at the governance layer.
The numbers reflect institutional confidence. Fireblocks secures over $4 trillion in digital assets using MPC. The broader MPC market stood at $888 million in 2024 and is projected to reach $2.72 billion by 2034. BitGo, Coinbase Custody, Cobo, and Anchorage have all built institutional custody around MPC as the cryptographic core.
The Honest Caveats
Most commercial MPC implementations hold one key share on the vendor’s own infrastructure. Fireblocks holds a share. BitGo holds a share. Cobo holds a share. This creates partial custodial dependency whether the marketing calls the product “self-custody” or not. You’re trusting the vendor’s key share, their server uptime, and their security posture—a meaningful counterparty relationship, even if it’s cryptographically structured differently from a traditional custodian.
The second caveat is more important: MPC doesn’t protect against governance attacks. The Bybit attack vector works identically against any MPC setup. If the authorized participants in a signing threshold are shown a fraudulent transaction and approve it, the TSS protocol executes it correctly. Every cryptographic guarantee remains intact, every dollar leaves.
This is why institutional MPC must be paired with a policy engine. Fireblocks’ enterprise policy engine evaluates transactions against more than 30 conditional parameters before signing is permitted: spending limits, counterparty whitelists, time-of-day restrictions, anomaly scoring. Consumer MPC products don’t have this. The difference between consumer-grade and institutional-grade MPC isn’t the cryptography—it’s the governance architecture built on top of it.
The 2025 Regulatory Inflection: Three Shifts That Changed Qualified Custody
While the industry processed the Bybit shock, three regulatory events in 2025 transformed the institutional custody landscape in the US. All three are directly relevant to any institution making custody decisions today.
SAB 121 repeal, early 2025. The SEC’s Staff Accounting Bulletin 121 required banks that custody crypto for clients to list those assets as liabilities on their own balance sheets—making bank-backed crypto custody economically prohibitive. The repeal removed that barrier entirely. Traditional banks can now compete in crypto custody without an artificial capital penalty. JPMorgan and other major institutions are evaluating custody services they couldn’t have justified offering before.
SEC no-action letter, September 30, 2025. The SEC’s Division of Investment Management confirmed that state-chartered trust companies can serve as “banks” under the Advisers Act and the Investment Company Act of 1940. Registered investment advisers and funds can now treat state trust companies as qualified custodians for digital assets with clear legal standing. The gray area that had persisted for years resolved in a single document.
NYDFS guidance update, September 30, 2025. The New York Department of Financial Services updated its virtual currency custody guidance with four explicit requirements: segregation of customer assets from corporate assets, separate accounting for customer holdings, sub-custodian disclosure requirements, and enhanced customer notifications. NYDFS already operated one of the most demanding crypto regulatory frameworks in the world. This update sharpened it further.
The market moved quickly. The OCC granted BitGo a federal trust charter in December 2025—giving it the same regulatory standing as a nationally chartered bank for custody purposes. Fireblocks Trust had already received NYDFS approval in 2024. Gemini went public on Nasdaq in September 2025 (ticker: GEMI) at a $3.3 billion valuation, holding both an NYDFS trust charter and MiCA authorization for EU operations. Anchorage has held a federal charter since January 2021.
The core takeaway: Qualified crypto custody in the US went from a regulatory gray area to a clearly structured category in H2 2025. The comparison for institutions shifted from “regulated risk vs. no risk” to regulated, examined, insured risk vs. self-managed risk with all the operational overhead that entails.
Three Institutions, Three Frameworks
The most important part of any custody analysis is usually the part that’s missing: different institutions have different right answers. Here are the possible three.
Persona 1: The Crypto-Native Fund ($10M–$500M AUM)
Primary concerns: counterparty risk, DeFi access, operational flexibility, LP reporting requirements.
Recommended architecture: institutional MPC wallet for operational and DeFi-accessible assets, with hardware cold storage for long-term reserves. A qualified custodian may be required regardless of technical preference—if LP agreements specify “qualified custody,” that’s a contractual obligation, not a technology choice.
The key question for MPC provider selection: which party holds a key share, and what happens operationally if that provider experiences downtime or insolvency? Understand the custodial dependency before signing.
Watch for: insurance coverage that explicitly addresses MPC key share compromise—a different risk category from traditional cold storage coverage, with different policy language.
Persona 2: The Corporate Treasury (Non-Crypto-Native Company)
Primary concerns: security, accounting treatment, insurance coverage, regulatory defensibility, board approval.
Recommended architecture: third-party qualified custodian with an NYDFS trust charter or OCC federal charter. For a treasury team that didn’t hire cryptographic engineers, the operational overhead of self-custody—key generation ceremonies, hardware management, recovery drills, geographic distribution of key material—is incompatible with a finance function that has other priorities.
The SAB 121 repeal makes bank-backed custody economically competitive. The September 2025 SEC no-action letter makes state trust companies unambiguous qualified custodians. Insurance and fiduciary protection matter more than control at this scale.
The key question: does the custodian’s insurance explicitly cover the specific perils relevant to your holdings? Commercial crime insurance and specific crypto custody insurance are different products with different exclusions. Read the policy before comparing headline coverage amounts.
Persona 3: The Crypto Product Company (Exchange, DeFi Protocol, Neobank)
Primary concerns: transaction throughput, operational speed, compliance coverage, customer asset protection.
Recommended architecture: tiered. MPC hot wallets for operational liquidity—with a full policy engine including velocity controls, counterparty whitelisting, and transaction simulation. A warm wallet layer for medium-term reserves. A third-party qualified custodian or hardware cold storage for customer asset reserves not required for daily operations.
Minimum audit baseline: SOC 2 Type II for the custody platform. Proof of reserves is becoming mandatory in an increasing number of jurisdictions and will continue to expand. Build for the audit requirement before it becomes compulsory.
Five Questions Every Institution Should Answer First
Before signing a custody agreement or committing to a self-custody architecture, answer these five questions.
1. Who actually holds the keys—and what share does the vendor control? “Self-custody MPC” and “third-party custodian” describe meaningfully different risk profiles. Most commercial MPC providers hold one key share on their own infrastructure. Understand the custodial dependency before claiming full self-custody in your risk documentation.
2. What is the provider’s regulatory status? An NYDFS trust charter, OCC federal charter, MiCA CASP authorization, and “registered in Wyoming” describe very different levels of regulatory oversight, capital requirements, and consumer protection. Ask which specific license applies—and what examination requirements come with it.
3. Does the insurance coverage match the actual risk? Ask for specific perils covered, not the headline coverage amount. Commercial crime insurance, specific crypto custody insurance, and staking slashing protection are different products with different exclusions. Confirm which applies to your assets and what the carve-outs are.
4. What is the transaction simulation and policy enforcement layer? The Bybit hack was a governance layer failure, not a cryptographic one. Before selecting any custody solution, run a live demo of how a suspicious transaction gets flagged. What triggers a block? What requires human review? What happens when a policy parameter is breached? A custody platform without this layer isn’t institutional-grade, regardless of the cryptography underneath.
5. Has the recovery procedure actually been drilled? Key refresh ceremonies, recovery quorums, and emergency access procedures are only useful if they’ve been tested under realistic conditions. Ask for documentation of a drill—not a description of the procedure, but evidence it was executed. A recovery procedure that exists only in a runbook isn’t a recovery procedure.
The Quantum Footnote
Brief, but worth flagging for institutions making long-term infrastructure bets.
NIST published post-quantum cryptography standards in 2024. ECDSA—the signing algorithm underlying virtually all crypto wallets and custody infrastructure—is theoretically vulnerable to a sufficiently powerful quantum computer. Migrating custody infrastructure to quantum-resistant signature schemes takes years to plan, approve, and execute across institutional infrastructure.
This isn’t a 2025 operational risk. It’s a 2030–2035 planning question. Forward-looking custodians are already building post-quantum migration roadmaps aligned with NIST’s PQC standards. When evaluating custody providers, “what is your post-quantum readiness roadmap?” is a reasonable addition to the procurement checklist. The infrastructure you choose today will still be operating when the answer matters.
Custody Is an Architecture Question, Not a Product Decision
The Bybit hack proved that cryptographic sophistication alone isn’t enough. Thirteen layers of multisig cold storage don’t protect you if the interface your signers use is compromised. The lesson isn’t to abandon multisig—it’s to treat custody as a systems design problem with cryptographic, governance, and UX dimensions that must all be addressed.
The 2025 US regulatory inflection proved that qualified custody now has clear standards, real competition, and genuine fiduciary protection that self-custody can’t replicate. The comparison for institutions is now “regulated, examined, insured risk” vs. “self-managed risk with all the operational overhead that entails.”
The right custody architecture depends on who you are. A crypto-native fund, a corporate treasury, and a crypto product company each have a different answer—and treating those answers as interchangeable is where institutional custody decisions go wrong.
The institutions that get this right in 2025–2027 will treat custody as infrastructure design: not a product to purchase, not a checkbox to tick, but a systems-level decision that determines what happens when—not if—something goes wrong.
Disclaimer: Please note that the contents of this article are not financial or investing advice. The information provided in this article is the author’s opinion only and should not be considered as offering trading or investing recommendations. We do not make any warranties about the completeness, reliability and accuracy of this information. The cryptocurrency market suffers from high volatility and occasional arbitrary movements. Any investor, trader, or regular crypto users should research multiple viewpoints and be familiar with all local regulations before committing to an investment.
