Where Are My Coins? — Wallets, Exchanges, and Who Holds the Key

Irreversible Transfers

Reviews of mobile apps used for sending and receiving coins feature this belief from some users. Transfers are free, sending to the wrong place can be undone, and you can cancel anytime. All three are false. This is the result of reading 6,859 reviews dealing with the user experience out of 45,821 reviews across five widely used apps (DOI: 10.1145/3411764.3445407).

Some of these misunderstandings come from trying to understand coins by relying on familiar payment methods, which leads to unmet expectations. Both first-time and long-time users got lost. The researchers noted that this confusion goes beyond mere frustration and drop-off, potentially leading to dangerous mistakes and the irreversible loss of money. They therefore recommended app designs that reduce misunderstandings and help first-time users (DOI: 10.1145/3411764.3445407).

Irreversibility is a property of coin transactions themselves. This is because there is no separate administrator to erase a transfer once it is written in the ledger. Scammers have exploited this property (DOI: 10.1145/3155808).

When you send money through a banking app, the record of that money is held by the bank. Coins are different. This article asks two questions. Where are my coins? Whose hands hold the power to move them? It does not recommend where to put them, nor does it deal with prices. How the law protects deposited money varies by country and system, so we only write up to the scope covered by the evidence.

What Is in Your Wallet Is a Key

As we saw in "The Right to Write a Line in the Ledger", a coin is literally the line written in the ledger that says "how much for whose share." That ledger is a notebook where many witnesses share identical copies, and this notebook is the blockchain. My line is in that notebook, not inside my mobile phone.

What is inside the wallet app on your phone are a public key and a private key, and the value of the coin itself is not inside (DOI: 10.1109/icccsp49186.2020.9315193). The two are a pair as seen in "Keys and Locks". The private key is the key that only the owner keeps and never sends anywhere, and the public key is the lock put out for anyone to use. A wallet is a tool that creates and holds a private key, and uses that key to make the signature that moves my line.

In coins like Bitcoin or Ethereum, what governs a share is the private key. When passing a share to someone else, the owner signs a new transaction and moves it to the recipient. Because of this, the key becomes an attractive target for attack, leading to a succession of cases where millions of dollars worth of cryptocurrency were stolen (DOI: 10.1145/3319535.3354236). The person who can sign is the person who can move it. Having a coin means holding the key to move that line in the notebook.

A recovery phrase is a line written in a few human-readable words so that the private key can be recreated. Anyone holding these words can enter that wallet from a new device. This is the way an owner who lost their phone revives their wallet, and it is also the way a stranger who gets their hands on the words enters the wallet.

Biometric lock screens like fingerprints or facial recognition set on a phone do not block someone else from getting in. That is because a stranger who gets their hands on the words enters from their own new device. To patch this gap, there are design proposals to use biometric information as a second authentication measure (DOI: 10.1145/3743093.3771660). A recovery phrase is a copy of the key.

If You Hold the Keys Yourself

Self-custody is a method where the owner holds the private key themselves, and the sole responsibility of protecting that key and the share hanging on it rests entirely on the owner. In such wallets, no third party steps in and the owner controls their share (arXiv:2409.15389, preprint). Blockchain shifts the responsibility of protecting assets to the user, requiring that person to manage their private key (DOI: 10.1145/3357236.3395535).

Because control belongs to the owner, the responsibility for losing it also belongs to the owner. If you lose both the private key and its copy, the recovery phrase, that share is lost irretrievably (DOI: 10.4324/9781003782216-16). If either one remains, it can be restored, but if both are lost, there is no place to retrieve it for you.

The fact that asset security depends on the owner guarding the key is the same whether it is an individual carrying daily spending money or a company moving large assets. There are many funds that have vanished because keys were lost, and because attackers got their hands on keys (DOI: 10.4230/oasics.tokenomics.2021.4). The common wallets of today that require keeping private keys as words also come with inconveniences. They are complicated to handle, awkward to keep safe, and do not include a separate feature to recover accounts (DOI: 10.1109/icbctis64495.2024.00017, design proposal paper).

Attackers can probe vulnerabilities in any form of wallet to steal money. The gap between possible countermeasures and those actually used is also wide (DOI: 10.1145/3596906). A widely used method to counter attacks targeting keys is to split the wallet in two and use them separately (DOI: 10.1145/3319535.3354236).

A hot wallet is a wallet that is always connected to the network. A cold wallet is a wallet that keeps the private key while disconnected from the network. Attacks coming across the network struggle to reach a key separated from the network. Therefore, the private key is kept on the cold side, and the hot side handles tasks that must always stay connected to the network.

When you hold them yourself, the power to move your share and the responsibility for losing your share come together to one person.

When You Hand Over the Keys

People usually get coins like Bitcoin in their hands through an exchange. An exchange is a place that trades coins for money or other coins (DOI: 10.1145/3155808). In that process, it is common not to touch the blockchain at all (DOI: 10.1109/ecrime57793.2022.10142141). This is because the records of buying and selling inside an exchange remain in a separate ledger kept by the exchange itself, and are not written in the notebook shared among witnesses.

About 99% of cryptocurrency trades happen in places like this, and many people leave their assets in their exchange accounts even after buying them (DOI: 10.1002/for.2846). At this time, the exchange holds the customer's coins on their behalf. A 2015 paper described this setup by noting that exchanges operate like banks (DOI: 10.1145/2810103.2813674). My coins left on an exchange do not exist as my line in the notebook, but as a number written in the exchange's ledger. The key backing up that number is held by the exchange.

Custodial storage is a method where a private key is held by a third party such as an exchange, which approves and signs the customer's transactions (DOI: 10.1109/icbc56567.2023.10174959; arXiv:2409.15389). Because it spares you the burden of handling the private key directly, it becomes easier to use. In return, three risks enter the picture: the risk of having to trust the place you entrusted, the difficulty of looking inside it, and the risk of the entrusted assets being misused (DOI: 10.4324/9781003782216-16).

When a single entity handles both trading and custody, the interests between the two can clash (DOI: 10.4324/9781003782216-16). This means that if the entity holding the customer's coins also runs the trading directly, what is good for the customer and what is good for the exchange can diverge.

When you hand custody over, you shed the burden of handling the keys yourself, and instead come to rely on whether the custodian guards them well, is honest, and does not close its doors.

The blockchain put forward a ledger that requires no trust in anyone, but users sought out and created centers of trust for themselves. Well-known exchanges were such centers, and the newer or less tech-savvy a user was, the more they preferred such intermediaries (arXiv:2505.06661). This is the finding of a 2025 preprint that looked together at interviews with 13 stakeholders and about 3,000 online bulletin board discussions.

The authors cite two states of mind as the reason. The desire for a place to hold accountable when things go wrong, and the reluctance to shoulder the full responsibility of self-custody. They read this as a tension between decentralization in theory and in practice. It means that trust in centralization stubbornly survives in an environment that was supposed to require no trust (arXiv:2505.06661).

What people put on the scale when choosing a wallet or an exchange are three risks that cause losses: one's own mistake, the betrayal of the trusted party, and someone's malicious attack. The researchers build a model where the estimation of risks and the intended use together guide the choice, and discuss design implications (DOI: 10.1145/3357236.3395535). Since it is a 2020 study that interviewed 10 users, the framework comes from ten people, but this article also uses these three as its yardstick.

Whichever side you pick, human misunderstanding remains. The researchers of a qualitative study targeting 29 users view it this way. Current tools like wallets and exchanges fail to fend off the threats brought on by user misunderstandings. Consequently, people frequently fail to handle private keys safely, and assume they are anonymous even when they are not. The researchers put forth advice tailored to how users understand things (DOI: 10.60882/cispa.24613239).

When the Custodian Closes Its Doors

When an exchange closes its doors, customer balances sometimes vanish with it. Following 80 Bitcoin exchanges established between 2010 and 2015 through to the period after, nearly half—38 of them—shut down, and in some cases, customer account balances disappeared. While fraudsters were sometimes to blame, that was not always the case (DOI: 10.1145/3155808).

Security breaches, meaning instances where an attacker broke into an exchange, appeared alongside closures. Twenty-five exchanges experienced breaches, and 15 of them subsequently closed. Broken down into three-month periods, the odds of closing within a given quarter were 13 times higher in a quarter that experienced a breach. The odds represent the likelihood of closing compared to the likelihood of not closing. This is an association showing the two appeared together (DOI: 10.1145/3155808).

The size of the exchange and the money it handled were also related to closure. For every doubling of trading volume, the odds of closure were 12% lower. Fiat currency is money issued by a state. Exchanges that mainly handled less common fiat currencies and faced only one competitor also closed less often (DOI: 10.1145/3155808).

Expanding the scope to 822 exchanges operating between 2010 and 2022, 40% closed down. Evidence linking breaches and closures was found again, though only partially. However, the factor with the strongest connection was how the exchange was operated (DOI: 10.1109/ecrime57793.2022.10142141).

Differences in operational methods looked like this: exchanges trading only coins without fiat currency had roughly 60% higher odds of closing than those handling both. The more coins an exchange handled, the less frequently it closed. Exchanges accepting U.S. customers closed faster, suggesting that the regulatory environment can affect an exchange's lifespan (DOI: 10.1109/ecrime57793.2022.10142141).

In a 2021 model predicting whether 238 exchanges would continue operating, whether they had suffered past breaches did not contribute significantly to the prediction. What contributed significantly were the duration of operation, trading volume, and security mechanisms. Security audits, cold wallet storage, and bug bounties (a system of rewarding people who find and report vulnerabilities) fell into this category (DOI: 10.1002/for.2846).

Hacking typically targets the weakest points of an ecosystem, and in Bitcoin, exchanges are one of them. Out of 36 Bitcoin exchange breaches analyzed in 2020, all but three were made possible by relatively lax security. The security level of exchanges fell short of other financial services (DOI: 10.1109/tnsm.2020.3046145).

There were also signs of improvement. Intrusions using stolen IDs and passwords—meaning break-ins that did not even require hacking—decreased. Both the amount of Bitcoin drained in a single breach and the number of exchanges closing after a breach declined. Although the overall security level improved, major flaws remained. Researchers view exchanges with lax security as intermediaries sitting between customers and the ledger, amplifying the risk they introduce to the ecosystem (DOI: 10.1109/tnsm.2020.3046145).

There are also instances where custodians misuse customer assets. Throughout the downturn known as the crypto winter, the misappropriation and misuse of assets—meaning siphoning entrusted assets away from their original place or diverting them elsewhere—were observed across the crypto-asset industry (DOI: 10.1093/cmlj/kmae010).

In November 2022, the FTX exchange collapsed. Examining transaction records on the ledger, researchers found that this exchange relied heavily on tokens it issued itself. It used those tokens as leverage—meaning to move sums larger than what it actually held—and misused them. That behavior worsened finances that were already fragile (DOI: 10.1016/j.physa.2023.129044, single-case analysis).

Researchers point to the collapse of another coin ecosystem as the decisive trigger for the downfall. Afterward, the exchange's liquidity—its capacity to convert assets into cash promptly—dropped sharply. A tweet posted by another major exchange appears to have provoked a market-wide reaction, actively accelerating the collapse. The paper's title also addresses how the exchange that posted that tweet solidified its dominance.

Researchers point out who benefited the most from this collapse and cite the trend of this space overall tilting toward centralization, where power concentrates in a few places. They emphasize that genuinely decentralized finance is crucial for a transparent digital economy. This emphasis concerns how the financial system should be structured (DOI: 10.1016/j.physa.2023.129044).

Custodians can be hacked, misuse assets, and close down, and hacking is not the only path to closure. Alongside this, there are records showing that security is improving.

Sharing the Keys

Multisignature is a method that places several keys and requires a set number of them to sign together for the funds to move. A common arrangement requires any two out of three keys. This way, if one is lost, the remaining two can still move the funds, and if one leaks, that alone is not enough to move them.

In a model comparing wallet designs by assuming the probabilities of losing or leaking each key, this "any two out of three" was the best design in many settings (DOI: 10.4230/oasics.tokenomics.2021.4). In the same model, the probability of failure generally decreased steeply as the number of keys increased. However, it heavily depended on whether the number of keys was odd or even.

Mixing keys with different probabilities of being lost or leaking was usually better, but not always. The authors leave measuring how often individual keys are actually lost and leaked through user studies as an open question. They write that these results guide users on how to construct their wallets (DOI: 10.4230/oasics.tokenomics.2021.4).

There was a gap in actual usage. There is another type of wallet that, like multisignature, does not keep keys concentrated in one place. These are wallets that split key pieces across multiple devices. Even though these wallets have better security guarantees than single-device ones, they were not used nearly as much up around 2023.

In a survey of 357 wallet users and subsequent qualitative analysis, there was a gap between the security users understood and the actual security guarantees. Depending on which threats they had in mind, the default settings they desired also differed. The researchers suggest that default settings must match the context of use, and propose design directions (DOI: 10.1145/3576915.3623218).

By sharing the keys, you can design systems to withstand a mistake, betrayal, or attack in any single place. The probabilities of losing and leaking still attach to each divided piece.

Who Checks the Custodian?

Multiple exchanges suffered catastrophic losses, and customers lost their savings forever. This happened up around 2015 (DOI: 10.1145/2810103.2813674). So customers want to know just one thing: does the exchange actually hold my share right now?

Proof of solvency is a method of using cryptography to show that an exchange actually holds enough coins to hand out to each customer account. Instead of the exchange just saying so with words, it means putting out a proof that anyone can check through calculation.

A method published in 2015 made it so this proof could be done without exposing the exchange's private business. The exchange does not have to reveal the addresses holding its coins, its total holdings or liabilities, or customer information either. An accompanying extension prevents exchanges from conspiring to cover up each other's losses. The researchers implemented this method and report that, even for large exchanges with millions of customers, the calculation time and proof size are practical (DOI: 10.1145/2810103.2813674, proposal and implementation). Whether exchanges actually use this method is not in its abstract.

Even if the ledger is public, checking does not happen automatically. Virtual asset service providers are businesses like exchanges that swap and hold coins. Looking at 24 virtual asset service providers registered with the Austrian Financial Market Authority in 2024, the coins they held were not yet subject to systematic auditing procedures. The financial services they provide were closer to brokers, currency exchange offices, and funds than to banks (DOI: 10.1080/00036846.2024.2396640).

The researchers put forward a method of measuring transaction flows on the ledger and comparing them against financial statements, and applied it to four places. The two data sources matched only partially. From this mismatch, the researchers found gaps in the data collection process, and together proposed a strategy to fill those gaps and move toward more systematic and reliable automated evaluation (DOI: 10.1080/00036846.2024.2396640). This means that based on public ledgers alone, we still cannot neatly reconcile whether the deposited share is fully there.

The scope of protection provided by law varies by country and system. The system covered by the sources for this article is the European Union's Markets in Crypto-Assets Regulation. A 2024 law paper views this regulation as focusing on having custodians organize and govern themselves soundly, and preventing customer assets from being reused in the firm's own accounts. At the same time, it assesses that the mechanisms to protect the assets themselves when a custodian, a third party, or an issuer runs into trouble are weak (DOI: 10.1093/cmlj/kmae010). This means the fence to protect customer assets precisely when the custodian shakes is weak.

A chapter dealing with custody lists regulatory currents such as proof of reserves, asset segregation (keeping customer assets separate without mixing them with company assets), licensing, and insurance. It then positions custody as the foundational layer that allows institutions to enter the market and the market to function properly (DOI: 10.4324/9781003782216-16). Whether that proof of reserves means the same thing as the proof of solvency earlier is not in the chapter's abstract.

"Keys and Locks" left behind the question: who guards the issuer of the name tag—the entity that guarantees whose lock this is? "What Is a Blockchain" and "The Right to Write a Line in the Ledger" noted that if a trustworthy place already exists and there is a way to verify that trust, the rationale for deliberately bringing in a device like a blockchain is weak. When you leave your keys with an exchange, a trustworthy place comes back into the picture, and the same question returns.

There are several ways out there to check whether a custodian really holds the customer's share, and each still comes with conditions attached.

One Person, Two Places

Now we follow a single person. This person splits their coins between two places. Some are kept in a mobile wallet, with the recovery phrase written on paper and put in a drawer. The mobile wallet is a hot wallet connected to the network, and this person holds the key directly. The rest are kept in an exchange account, and the exchange holds the key.

We apply three risks—namely, mistakes, betrayal, and attacks—in turn to both places. The names of the three risks came from ten interviews, and that study gives only the names. This article does the work of matching them to the two places.

First is mistakes. If the phone is lost, having the paper in the drawer lets you restore the wallet on a new device. If both the phone and the paper are lost, the portion on the phone side is lost beyond recovery (DOI: 10.4324/9781003782216-16). On the exchange side, the private key to be lost is never in this person's hands from the beginning, and the burden of handling keys is taken off by the exchange.

Still, mistakes remain in both places. A transfer sent with a wrong recipient address cannot be canceled once it is written to the ledger. It is the same whether it was sent from a mobile phone or sent out from an exchange account. This is because irreversibility is a property of the transaction itself (DOI: 10.1145/3155808). The researchers of the 29-paper study seen earlier also viewed both wallets and exchanges as failing to fend off threats caused by users' misunderstandings (DOI: 10.60882/cispa.24613239).

Second is betrayal. Anyone who gets their hands on the words on the drawer's paper can enter this wallet from a new device. It is the same even if it was someone trusted. On the exchange side, there is the risk of misusing entrusted assets, and the misappropriation and conversion of assets were actually observed in this industry during the crypto winter (DOI: 10.1093/cmlj/kmae010).

If this person's exchange is in the European Union, its regulations prohibit reusing customer assets in its own accounts. Still, a 2024 law paper evaluates that the mechanisms protecting the assets themselves when an exchange falls into distress are weak. What protects this person elsewhere is not covered by the evidence in this article.

Third is attacks. A wallet in any form can have attackers probe its weaknesses, and this person's mobile wallet attached to the network falls under that. An exchange is one of the weak spots that hacking targets. Among 80 Bitcoin exchanges established between 2010 and 2015, 25 suffered breaches, and 15 of those later shut down.

However, this is a correlation where breaches and closures appeared together. In an analysis with a broader scope, the factor most strongly linked with closure was operational practices, such as whether they handled only coins. In the 36 cases seen earlier, breaches were possible due to loose security in all but three. Even so, the number of exchanges closing down after a breach and the amount escaping in a single breach were decreasing.

Outside the three risks, this article adds one more. An exchange can shut down even without betrayal or attacks, and there were cases where balances vanished as it shut down. It was sometimes the fault of scammers, but not always (DOI: 10.1145/3155808).

There was no place free of risk in either place. On the mobile side, the three risks point toward the key held by this person and the paper in the drawer. On the exchange side, while the worry of losing the key decreases, betrayal and attacks target the honesty and security of the exchange, and one more risk of shutting down is added. This person's own misunderstandings and mistakes remain in both places.

Whether you hold the key directly or entrust it, the risk does not disappear; what changes is which risk is borne by whom. This sentence is this article's synthesis of the preceding evidence. Which side to view more heavily depends on how this person weighs the three risks and what they intend to use the coins for. In the ten interviews, people actually chose tools that way (DOI: 10.1145/3357236.3395535).

One more question remains on the exchange side. Does the exchange really hold this person's share right now? There is no single answer to this question either. Proof of reserves is a proposal that researchers even demonstrated by implementing it. The method of matching flows on the ledger against financial statements had gaps in the data-gathering process when tested. A 2024 law paper evaluated that the EU's regulations are weak at protecting the assets themselves.

There are also two ways to bypass this question. One is a design where the key is split and held among multiple parties, which was not widely used until around 2023. The other is the path of holding the key directly, which returns the risks of mistakes, betrayal, and attacks targeting the phone back to this person.

Where to Look Further

What it means for a promise written in a ledger to be executed exactly as written is covered by the final piece of this series, "What Is a Smart Contract? — A Promise Everyone Runs the Same Way, Exactly as Written" (/articles/smart-contract-primer).

How people find themselves looking for a center to trust once again does not end at exchanges. "Is Blockchain Really Decentralized — From Mining Pools to MEV, Where Power Reassembles" (/articles/blockchain-decentralization-debate) measures where the right to write in the ledger gathers.

The decision to make anyone able to see what a trusted place did in a public record, instead of picking somewhere to trust, is covered by "What Does the Padlock in Your Address Bar Actually Guarantee? — Instead of Picking a Trustworthy Authority, the Web Decided to Write Every Issuance Down" (/articles/certificate-transparency). It is an answer given in a different place to questions like the proof of solvency in this article.

The exchange defined in this article appears without definition in several parts of "Things That Collapsed the Moment They Made Money". Cases where a permission left in the code was the problem are handled by "A Rug Pull Is Not a Hack — It Is a Permission" (/articles/rug-pull-anatomy), cases where the conditions written to maintain value by themselves collapsed by "The Code Did Exactly What It Promised — What Algorithmic Stablecoins Guaranteed, and What They Could Not" (/articles/algorithmic-stablecoin-collapse), and the work of bringing in prices from outside the ledger by "A Smart Contract Cannot See the Outside World — The Moment You Have to Trust Whoever Tells It the Price" (/articles/defi-oracle-manipulation).

In the same bundle are "A Bridge Does Not Inherit the Chain's Guarantees — Why Cross-Chain Bridges Keep Getting Drained" (/articles/cross-chain-bridge-hacks), which deals with the spaces connecting ledger and ledger, and "The Power to Change the Rules Is Also for Sale — When Governance Itself Becomes the Attack Surface" (/articles/dao-governance-attacks), which deals with buying votes to change the rules.

Source Papers

Design proposals, single-case analyses, preprints, and models are explicitly marked for their rigor in the text.

• "The U in Crypto Stands for Usable: An Empirical Study of User Experience with Mobile Cryptocurrency Wallets" (2021) — DOI: 10.1145/3411764.3445407 — Qualitative analysis of 6,859 UX-related posts out of 45,821 reviews of top 5 mobile wallet apps. Both new and experienced users struggle, which might lead to dangerous mistakes and irreversible losses (might). Misconceptions that transactions are free, irreversible, and can be cancelled at any time lean on traditional payment understandings. Design recommendations to help new users.
• "Cryptocurrency Wallet: A Review" (2020) — DOI: 10.1109/icccsp49186.2020.9315193 — Wallets store public and private keys, not actual currency values. Overview of software, hardware, and paper wallet classifications along with multi-currency wallet features.
• "A Blockchain Wallet Scheme with Multi-Factor Authentication Based on Distributed System" (2024) — DOI: 10.1109/icbctis64495.2024.00017 — A wallet is a technology that generates, stores, and uses private keys to manage assets. Traditional wallets help store keys using mnemonics, but manipulation is complex, safe storage is inconvenient, and account recovery is not supported. Design and implementation of a secret-sharing-based multi-factor authentication wallet (design proposal).
• "A Formal Treatment of Deterministic Wallets" (2019) — DOI: 10.1145/3319535.3354236 — Users control funds via secret keys, and owners transfer them by signing new transactions. This makes keys an attack target, resulting in millions of dollars in theft cases. As a countermeasure, the first comprehensive security model for widely used hot (always connected) and cold (disconnected, storing secret keys) wallets along with a provably secure ECDSA-based design.
• "Biometric-Based Recovery Mechanism for Blockchain-Based Wallet" (2025) — DOI: 10.1145/3743093.3771660 — In non-custodial wallets, the responsibility for key protection lies with the user, and recovery is usually impossible if compromised. Device biometrics do not stop an attacker who has obtained the seed phrase from accessing a new device. We propose remote authentication using biometric information as a second authentication factor to reduce the risk of seed phrase theft (design proposal).
• "Custodial and Non-Custodial Wallets" (2024, preprint) — arXiv:2409.15389 — Non-custodial wallets give the owner sole control over private keys, carrying sole management and protection responsibility. Custodial wallets are managed by a third party like an exchange. Summarizes differences, features, and risks.
• "Don't lose your coin! Investigating Security Practices of Cryptocurrency Users" (2020) — DOI: 10.1145/3357236.3395535 — Semi-structured interviews N=10. Blockchains shift the responsibility of asset protection to users and demand key management. Tool choice is driven by weighing and balancing three risks that lead to loss: human error, betrayal, and malicious attack. A model where risk estimation and intended use drive choice, along with design implications.
• "Custodial Wallets and Custodians" (2026, book chapter) — DOI: 10.4324/9781003782216-16 — Custodial wallets ease the burden of key management to make them user-friendly, but introduce risks of trust, transparency, and asset misappropriation. Decentralized wallets give direct control, but losing private keys or mnemonics means irreversible loss. Discusses conflicts of interest from combining trading and custody, cold/hot separation in institutional custody, expectations and risks of decentralized custody, and regulatory trends like proof of reserves, asset segregation, authorization, and insurance, positioning custody as the foundational layer for institutional participation and market integrity.
• "On Cryptocurrency Wallet Design" (2022) — DOI: 10.4230/oasics.tokenomics.2021.4 — Security relies on the owner's key protection, applying to both individuals and large asset managers. Many losses stem from key loss and theft. In a model where each key has probabilities of security, loss, leakage, and theft, failure probability drops exponentially with the number of keys but depends heavily on parity; heterogeneous keys are usually (though not always) superior, with a requirement of any 2 out of 3 keys optimal in many settings. Open questions such as user studies on key defect probabilities, and practical information on optimal design.
• "User Mental Models of Cryptocurrency Systems - A Grounded Theory Approach" (2020) — DOI: 10.60882/cispa.24613239 — Qualitative study N=29. Misconceptions affecting security and privacy. Current tools (wallets and exchanges) fail to prevent the threats triggered by these misconceptions; users frequently fail to manage private keys securely and believe they are anonymous when they are not. Actionable advice grounded in mental models.
• "Cryptocurrency Exchange Closure Revisited (Again)" (2022) — DOI: 10.1109/ecrime57793.2022.10142141 — Most people buy cryptocurrency through exchanges and often never touch the blockchain. 40% of 822 exchanges established between 2010 and 2022 have closed. Some evidence links breaches to closures, but the strongest effect comes from operational practices: exchanges trading only coins have an ~60% higher odds of closure, more listed coins lower closure odds, and allowing US customers leads to faster closure — suggesting regulatory environment impacts.
• "Revisiting the Risks of Bitcoin Currency Exchange Closure" (2018) — DOI: 10.1145/3155808 — Exchanges exchange Bitcoin for fiat currency. Out of 80 exchanges founded between 2010 and 2015, 38 closed, sometimes with balances vanishing. Fraudsters are not always the cause. 15 out of 25 breached exchanges closed. The odds of closure in the quarter of a breach are 13 times higher (correlation). Odds drop 12% for every doubling of trading volume, and exchanges centered on less common fiat currencies with at most one competitor are less likely to close.
• "Cryptocurrency exchanges: Predicting which markets will remain active" (2021) — DOI: 10.1002/for.2846 — About 99% of trades take place on exchanges, and many investors leave their assets in exchange accounts, exposing them to closure risk. Machine learning prediction using 238 exchanges and 8 characteristics (best model Random Forest, test accuracy 86.1%). Operational lifespan, trading volume, security audits, cold storage, and bug bounties rank highest in importance, whereas past breaches do not rank at the top.
• "Provisions" (2015) — DOI: 10.1145/2810103.2813674 — Exchanges hold customers' Bitcoins in custody like banks. Several exchanges suffered catastrophic losses where customers permanently lost their savings. Proof of solvency cryptographically proves control of reserves to settle customer accounts. Proposes and implements a method that hides addresses, total holdings, liabilities, and customer information while scaling to prevent collusion among exchanges, practical even at the scale of millions of customers.
• "Crypto Assets Custody: Taxonomy, Components, and Open Challenges" (2023) — DOI: 10.1109/icbc56567.2023.10174959 — Custodians store private keys and authorize and sign transactions. They must operate efficiently and securely without loss or theft. A survey, taxonomy, and open challenges of custody solutions.
• "Centralized Trust in Decentralized Systems: Unveiling Hidden Contradictions in Blockchain and Cryptocurrency" (2025, preprint) — arXiv:2505.06661 — 13 stakeholder interviews + about 3,000 Reddit discussions. Users seek out and establish centralized trust anchors like well-known exchanges. This is driven by the need for accountability and reluctance to bear the full responsibility of self-custody. New and non-technical users prefer intermediaries more. The tension between theoretical decentralization and reality, and the persistence of centralized trust.
• "Cyber Security Threats to Bitcoin Exchanges: Adversary Exploitation and Laundering Techniques" (2020) — DOI: 10.1109/tnsm.2020.3046145 — Analysis of 36 Bitcoin exchange breaches. All but three were due to relatively lax security. Security levels fall short of other financial services. Decreased use of compromised credentials, lower stolen amounts per breach, and fewer closures following breaches indicate an overall improvement in security levels, yet major flaws remain. Analysis of two laundering cases shows that exchanges with lax security amplify intermediary risk.
• "Crypto custody" (2024) — DOI: 10.1093/cmlj/kmae010 — Evaluating the custody regulations of EU MiCA against the backdrop of asset appropriation and misuse during the crypto winter. MiCA focuses on sound organization and governance of custodians and the prohibition of reusing client assets on their own account ("institutional resilience"), but "asset resilience" is weak when custodians, third parties, issuers, or DeFi run into trouble.
• "FTX's downfall and Binance's consolidation: The fragility of centralised digital finance" (2023) — DOI: 10.1016/j.physa.2023.129044 — Causes and consequences of the November 2022 FTX collapse. On-chain data shows FTX heavily relied on and misused its own token FTT as leverage, worsening its fragile financials. The Terra-Luna collapse was a decisive turning point for liquidity decline, and Binance tweets appear to have triggered a systematic response that accelerated the collapse. Points out the profiting actors and centralization trend, emphasizing the importance of genuine decentralized finance.
• "Uncovering Impact of Mental Models towards Adoption of Multi-device Crypto-Wallets" (2023) — DOI: 10.1145/3576915.3623218 — Multi-device wallets offer better guarantees than single devices, yet adoption does not scale proportionally. Survey of 357 people and follow-up qualitative analysis: a gap between mental models and actual guarantees, requiring contextualization since preferred defaults vary depending on threat models. Suggests design directions.
• "Assessing the solvency of virtual asset service providers: are current standards sufficient?" (2024) — DOI: 10.1080/00036846.2024.2396640 — 24 VASPs registered with the Austrian FMA, handling 2 billion euros in annual inflows/outflows and 1.8 million customers; their services resemble brokers, currency exchangers, and funds rather than banks. Held cryptoassets are not yet subject to systematic audits. Comparing on-chain flows and financial statements for four firms shows partial matches. Points out gaps in data collection and proposes systematic and automated assessment strategies.
• "Security Aspects of Cryptocurrency Wallets—A Systematic Literature Review" (2023) — DOI: 10.1145/3596906 — In all wallets, attackers can exploit weaknesses to steal money. 6 categories of attack vectors. A wide gap between possible countermeasures and actual adoption, with directions for future research.