Crypto Asset Custody Guide: How Should Institutions Choose a Safe and Reliable Crypto Asset Custody Solution?

On February 21, 2025, a multi-signature cold wallet belonging to Bybit, a leading global exchange, was breached, with losses exceeding $1.4 billion (401,347 ETH) — making it the largest theft in cryptocurrency history. The attacker was the Lazarus hacking group, widely believed to have North Korean state backing. Their method did not involve directly cracking the private key or the multisig algorithm itself, but rather compromising a Safe developer’s machine and injecting malicious script into the signing process, so that Bybit’s multiple signers saw what appeared to be a “normal transaction” on screen while actually signing a tampered contract call. After the signature was completed, the malicious code quietly switched the interface data back to normal, leaving no trace. What makes this incident especially worth institutions’ attention is that it bypassed not the cryptographic algorithm, but the assumption that “what the human eye sees equals what is actually being signed” — a risk the industry calls “blind signing.”

A year later, today, the same core problem has recurred in a different form. Over just a few days starting July 30, 2026, more than 5,200 addresses using Coldcard hardware wallets were attacked, with approximately 1,816 bitcoins (worth about $116 million) stolen. The cause was that firmware released in 2021 had inadvertently weakened the random number strength in the key generation process to just 40 bits, allowing attackers to recover private keys through brute force without physically accessing the device.

From Bybit’s multisig blind-signing vulnerability to Coldcard’s key generation flaw, the two incidents involved completely different attack surfaces, yet both lead to the same conclusion: the assumption that “self-custody equals security” does not hold. As a result, “crypto asset custody” — a topic once confined mostly to technical circles — has been pushed back onto the desks of institutional investors, exchanges, and compliance departments.

According to CoinDesk statistics, losses from security incidents in the crypto industry reached \$972 million in the first half of 2026, with 54.6% of those losses stemming from key management failures and compromised centralized systems, rather than smart contract code vulnerabilities themselves. This does not even include the Bybit incident, which alone exceeded \$1.4 billion — a historic-scale event on its own.

So the core conflict in crypto asset security has long since shifted from “is the code written correctly” to “how exactly should the private key be stored, and can the moment of signing actually be trusted.” This is the real reason crypto asset custody technology exists.

What Is Crypto Asset Custody, and Why Does It Determine Whether Digital Assets Live or Die

Crypto asset custody refers to the full lifecycle of secure management services covering the generation, storage, use, and recovery of digital asset private keys (or corresponding permissions). Blockchain’s “minimized trust” property means that once a private key is lost or stolen, the assets are almost impossible to recover — a complete departure from traditional financial systems, where banks and clearinghouses can freeze, trace, and intercept transactions.

Cumulative digital asset losses suffered by institutions due to cyberattacks have exceeded $14.7 billion (at historical market prices; at 2021 market prices, cumulative asset losses reach roughly $100 billion), with multiple attacks suspected to involve state-backed hacking groups. Safeheron categorizes common institutional security vulnerabilities into five types: insufficient sensitivity to phishing, hot wallets lacking security protection, weak protection of employee login authorization, vulnerabilities in the system itself, and tampering during the transaction process. These five categories remain the primary causes of security incidents in the industry today, and the Coldcard incident is, in essence, yet another version of “a systemic vulnerability in the private key generation process.”

Historical Attack Cases Show That Private Key Single Points Are Crypto Asset Custody’s Greatest Enemy

Tracing the industry’s major security incidents over the past decade or so in chronological order, nearly all of them point to the same root cause — single-point storage or use of the private key, or an untrustworthy signing process itself:

• Mt. GOX (2014): Losses of $473 million due to tampered transaction information;
• Bitfinex (2016): About 120,000 bitcoins stolen due to flaws in hot wallet private key management;
• Parity multisig wallet (2017): A smart contract vulnerability led to about 150,000 ether being frozen or stolen, with losses of about $30 million;
• Coincheck (2018): A single-private-key hot wallet was breached, with losses of about $534 million worth of XEM tokens;
• The “CryptoCore” crime group (2018–2021): Over three years, stole approximately $200 million in digital assets from multiple exchanges through phishing;
• KuCoin (2020): Private keys for hot wallets holding various digital assets were stolen, with losses of about $250 million;
• Bybit (2025): The largest single theft to date — the Safe{Wallet} multisig cold wallet was breached by the Lazarus hacking group through malicious tampering of the signing interface (the “blind signing” vulnerability), with losses exceeding $1.4 billion.

Over the past decade, attack methods have evolved from hot wallet private key leaks and smart contract vulnerabilities to tampering with the multisig signing interface itself, and loss scale has escalated from the tens-of-millions-of-dollars level to the billion-dollar level seen in the 2025 Bybit incident. But these cases all point to one conclusion: whether it’s a single-private-key hot wallet scheme, a traditional centralized custody model, or a seemingly more secure on-chain multisig scheme, as long as the private key appears in complete form at any stage, or the information a signer “sees” can be tampered with in the middle so that it differs from what is “actually signed,” there is a risk of being breached at a single point. This is exactly why institutional-grade crypto asset custody solutions have, in recent years, turned toward “Secure Multi-Party Computation (MPC)” architecture, and made “what you see is what you sign” a hard requirement for signing security.

MPC Technology Splits the Private Key Apart, Rather Than Locking It Up

Secure Multi-Party Computation (MPC) was first proposed in 1982 by Turing Award laureate and Chinese-American computer scientist Professor Andrew Yao, aiming to solve the problem of how multiple data holders who do not trust one another can compute collaboratively, such that no party obtains any information beyond the result they are entitled to. Applying this cryptographic theory to private key management means that throughout the entire chain from generation, to storage, to use, the complete private key in plaintext never appears on any single terminal — instead, “private key shares” held separately by multiple parties jointly complete the signature through collaborative computation.

Compared with on-chain multisig schemes, MPC offers several notable advantages:

1. It is not limited by asset type and can serve as a unified solution applicable to nearly all on-chain assets;
2. On-chain, it appears as a single-address transaction, with fees 30%–40% lower than on-chain multisig;
3. The number of signing parties can be flexibly adjusted, without being constrained by the signature-count limits of the protocol itself, as on-chain multisig is;
4. At the same time, it appears on-chain as an ordinary single-signature transaction, protecting the institution’s on-chain privacy, while still allowing every signing participant to be fully traced off-chain.

Safeheron’s Secure Custody Solution: MPC, a Trusted Execution Environment, and Zero-Trust Security Architecture

Safeheron‘s MPC self-custody product splits the full lifecycle of the private key into several key stages:

1. Private key generation: After the client initiates a creation instruction, the client and Safeheron each independently generate a private key share on their respective terminals, and generate the public key through an encrypted channel and the MPC protocol — the corresponding complete private key never appears on any terminal;
2. Private key backup: The client’s terminal exports its share, and only the client themselves can decrypt all the share ciphertexts. Even in the extremely unlikely event of a catastrophic platform failure, the client can still safely recover all assets using an open-source recovery tool;
3. Transaction initiation and approval: When any member initiates a transaction, the system verifies identity and checks it against preset policy engine rules — for example, “transfers over $1 million require approval from at least 2 people” or “weekend transactions require joint approval from an administrator and at least 1 member” — with multiple people within the organization reviewing the transaction’s amount, address, and purpose, guarding against human error and man-in-the-middle hijacking;
4. Signing and broadcasting: Once approved, the MPC protocol coordinates the client’s terminal and Safeheron’s terminal to jointly complete the signature computation — the original private key never appears on any terminal throughout the process — after which nodes distributed across multiple regions worldwide securely broadcast the signature to the chain.

Beyond this process, Safeheron also introduces a Trusted Execution Environment (TEE) to protect key private information and the signing program, so that even if a server is compromised or an insider intends harm, this private data cannot be taken or tampered with; this is paired with a “full-chain authentication” zero-trust security architecture, under which, if any link in the chain is verified as untrustworthy, the related behavior is automatically detected and blocked. The whitepaper particularly emphasizes the principle of “what you see is what you sign,” requiring that the transaction information a signer sees on their terminal must be completely identical to the information actually signed in the end, with no tampering allowed in between — this is precisely the direct solution to the vulnerability exposed in the Bybit incident, where the Safe{Wallet} signing interface was maliciously “swapped.” On the compliance side, Safeheron has integrated with leading anti-money-laundering service providers in the industry, and undergoes continuous security audits and penetration testing through third-party specialist organizations such as SlowMist and Kudelski Security; the relevant audits and security practices can be viewed publicly at the Safeheron Trust Center.

Safeheron states that, compared with the three other categories of solutions — traditional custody, contract-based multisig, and native on-chain multisig — the MPC path achieves full compliance across nearly 20 key security metrics, including private key generation security, private key storage security, private key usage security, private key leak-response mechanisms, private key recovery security, “what you see is what you sign,” and resistance to man-in-the-middle attacks. The other three categories of solutions (including the contract-multisig type used by Bybit) generally show clear shortcomings in dimensions such as private key generation/storage/usage security, “what you see is what you sign,” and resistance to man-in-the-middle attacks — which correspond precisely to the two root problems exposed by the Bybit incident (“signing information can be tampered with in the middle”) and the Coldcard incident (“insufficient security in the key generation process”), respectively.

The 2026 Regulatory Climate: Compliance Is Becoming a Hard Constraint on Custody Solution Selection

The industry’s security anxiety is unfolding in tandem with tightening regulation worldwide. After the U.S. GENIUS Act formally took effect, it established a clear framework for reserve asset custody and information disclosure for compliant stablecoin issuers. In July 2026, Taiwan also passed new legislation requiring crypto enterprises and stablecoin issuers to apply for approval from financial regulatory authorities under a stricter regulatory framework. This means that, going forward, whether for exchanges, OTC brokers, or asset management institutions, when choosing a crypto asset custody provider, beyond security itself, having robust anti-money-laundering capabilities, an audit trail mechanism, and transparent proof of fund storage will equally be an unavoidable compliance test.

Five Questions Institutions Should Focus On When Choosing a Crypto Asset Custody Solution

Taking into account the industry incidents and regulatory trends above, when evaluating crypto asset custody providers, institutions are advised to confirm at least the following:

1. Whether the private key has ever appeared in complete plaintext form at any stage of generation, storage, or use;
2. Whether the provider supports flexible, customizable multi-person approval policies, rather than a simple fixed multisig threshold;
3. Whether it has an open-source, verifiable private key recovery mechanism, avoiding a situation where the platform alone has unilateral control over the assets;
4. Whether it is connected to recognized anti-money-laundering and compliance service capabilities;
5. Whether it has been audited and penetration-tested by an independent third-party security organization, with verifiable audit records available.

For asset managers, venture capital firms, and family offices, custody solutions aimed at funds and asset management institutions typically also need to satisfy the dual requirements of multi-person approval and transparent oversight for investors; for OTC brokers and trading firms, the OTC and brokerage solution places greater emphasis on automated settlement efficiency in a high-frequency trading environment; for exchanges and payment service providers, the corresponding custody solution needs to balance automated deposit and withdrawal of user assets with the trust endorsement that third-party custody provides.

Conclusion

From Mt. GOX and KuCoin, to the largest incident in history in 2025 — Bybit — and on to the 2026 Coldcard hardware wallet incident, the forms crypto asset security incidents take keep changing, but the core conflict has never changed: as long as the private key exists in complete, single-point form at any stage, or “what is seen” and “what is signed” during the signing process can be separated, an attack surface exists. MPC, combined with a Trusted Execution Environment and zero-trust architecture, is essentially a cryptographic engineering means of fundamentally eliminating the private key single point, returning the question of “who can move the assets” to the true owners of those assets.

If you are evaluating crypto asset custody solutions for an institution, you are welcome to schedule a demo with the Safeheron technical team to learn more about custody architectures and security solutions tailored to your specific business scenario.