Crypto platforms lost about $1.18 billion to hacks between July and September 2026, across 107 incidents, according to Hacken’s quarterly report of 15 October as relayed by Cointelegraph. On 24 September the exchange Bitget lost about $388 million from its hot and warm wallets on 11 blockchains. The thief never had to break any of those chains: the withdrawals were forged inside Bitget’s own systems, and the networks simply recorded them as valid transactions.
That is the short answer to the question can a blockchain be hacked. The core rules of Bitcoin and Ethereum have never been broken by an attacker. Software bugs have been found in them and fixed, small blockchains have been taken over in so-called 51% attacks, and billions of dollars are stolen every year from exchanges, bridges, smart contracts and the people who hold the keys. The chain rarely breaks. The doors built around it do.
To read any hack headline, ask which layer was broken. Once you can place an incident in one of five layers, you know whether it says anything about the blockchain itself, and whether it says anything about where your own coins sit. For the scams aimed at you personally, the previous lesson of this path, our guide to avoiding crypto scams, goes into the red flags and where to report them.
Can a blockchain be hacked? Five layers, five different answers
A blockchain such as Bitcoin is a shared ledger that thousands of computers keep in step. Each block points to the one before it through a hash, so changing an old transaction would change every block after it, as explained in our lesson on how a blockchain works. Ownership is proved by a digital signature made with a private key. Breaking that layer would mean forging signatures or rewriting history against the will of the network.
Most of what the news calls a “blockchain hack” happens somewhere else. Wallet apps, exchanges, lending protocols and bridges are programs and companies that sit on top of the chain or next to it. They hold keys and run code, and people approve whatever they show on a screen. Each of them can fail without the chain failing.
| Layer | What can go wrong | A real example |
|---|---|---|
| 1. Protocol and consensus | Someone gains enough mining power or stake to reorder recent blocks | Ethereum Classic, July and August 2020: 807,260 ETC double-spent (Bitquery) |
| 2. Node software | A bug in the code that checks blocks lets an invalid transaction through | Bitcoin, 15 August 2010: a transaction created more than 184 billion BTC |
| 3. Smart contracts and bridges | A flaw in the contract’s logic, or stolen keys of the people who run a bridge | The DAO, 2016: over 3.6 million ETH drained (ethereum.org) |
| 4. Exchanges and wallet software | Attackers reach the systems that hold or sign with the keys | Bybit, 21 February 2025: about $1.5 billion (FBI) |
| 5. People | A victim types a seed phrase into a fake site or signs a malicious transaction | Ngrave app users, 17 October 2026: at least $2.56 million from about 70 wallets |
Why rewriting a large blockchain costs so much
On a proof-of-work chain, the history that counts is the one with the most computing work behind it. To change a past block, an attacker would have to redo the work of that block and of every block mined since, then overtake the honest miners who keep adding new ones. Satoshi Nakamoto’s 2008 white paper builds its security on that race. It assumes that honest nodes control most of the computing power, and it argues that an attacker who did gather that much power would rather use it to earn new coins: in its words, “he ought to find it more profitable to play by the rules”.
Ethereum has used proof of stake since 2022, and the defence works differently. Validators lock up ether to take part. According to ethereum.org’s page on attacks and defences, a group holding more than a third of all staked ether could stop the chain from finalising blocks, and an attacker would need far more to rewrite finalised history. Validators caught breaking the rules lose part of their stake, and as a last resort the community can coordinate outside the protocol and agree to follow an honest fork, leaving the attacker’s chain and stake behind.
Neither design makes an attack impossible. They make it expensive and, on a large network, likely to destroy the value the attacker was trying to steal. Even a successful majority attack has limits: it cannot produce a valid signature without the private key, so it cannot move other people’s coins. What it can do is reverse the attacker’s own recent payments, which is called a double spend, and refuse to include other people’s transactions.
Bitcoin’s two real scares: 2010 and 2018
People who ask whether Bitcoin has ever been hacked usually have one of two events in mind. Neither was a break of the cryptography. Both were bugs in the program that checks blocks, and both were fixed before they could do lasting harm.
August 2010: 184 billion bitcoins for three addresses
On 15 August 2010, block 74638 contained a transaction that created 184,467,440,737.09551616 bitcoins for three addresses, according to the Bitcoin Wiki’s account of the value overflow incident. Two of them received about 92.2 billion BTC each. The total supply was meant never to exceed 21 million.
The cause was arithmetic. Each output was a number so large that, when the software added the outputs together to check that they did not exceed the input, the sum wrapped around past the largest value the program could store and came out as a negative number. The check passed. Developers on the Bitcointalk forum spotted the strange block the same day, and the wiki says a new version of the client was out within five hours of the discovery.
Satoshi posted the fix as version 0.3.10, with instructions to shut down, replace the block files with a clean copy from before the bad block, upgrade and let the software download the rest. The patched software rejected any transaction whose outputs overflowed or exceeded 21 million BTC. That split the chain for about a day: unpatched nodes kept building on the bad block while upgraded ones built a new branch without it. The honest branch overtook the other at block 74691 on 16 August, the wiki says, and from then on every node treated it as the real history. The 184 billion coins never existed on the chain Bitcoin runs on today.
September 2018: the duplicate-input bug, CVE-2018-17144
Eight years later, a flaw in Bitcoin Core versions 0.14.0 to 0.16.2 could have let a miner create a block in which a transaction spent the same coins twice. Version 0.14 would simply crash on such a block, which is a denial of service. In versions 0.15 and 0.16, under certain conditions, the block would be accepted and the coins counted twice, which is inflation.
The Bitcoin Core disclosure of 20 September 2018 gives the timeline. The bug was reported on 17 September as a crash only. The same day, developer Matt Corallo found that it was also an inflation bug. A fixed release, 0.16.3, came out on 18 September with notes describing only the crash risk, to give operators time to upgrade before the worse half became public. By the full disclosure two days later, the developers believed over half of Bitcoin’s hash rate ran patched software, and they wrote: “We are unaware of any attempts to exploit this vulnerability.”
What the two bugs have in common
In both cases the signatures and hashes did exactly what they were meant to do. The weak point was ordinary code that counts amounts, found by people reading blocks and source code, and closed by an upgrade that node operators chose to install. That pattern keeps repeating. On 9 October 2026 the XRP Ledger disclosed an integer overflow bug that had sat in its payment engine since 2015, patched on 25 September with no sign of exploitation. On 14 October the privacy coin Pirate Chain found a validation bug that had let someone mint about 39.8 million ARRR, roughly a fifth of its supply.
Smaller chains and 51% attacks
A 51% attack is the blunt version of breaking layer one: rent or build more mining power than the rest of the network combined, mine a private chain in secret, then publish it so that it replaces the public one. Coins the attacker had sent to an exchange, and already traded or withdrawn, come back to the attacker’s own address on the new history. The exchange is left with a deposit that no longer exists.
Bitcoin Gold, a 2017 offshoot of Bitcoin with far less mining power behind it, went through this in May 2018. The attacker targeted exchanges, not individual users, and more than 388,000 BTG passed through the address the stolen funds were traced to, worth over $18 million at the time, BleepingComputer reported on 24 May 2018. “The cost of mounting an ongoing attack is high,” the Bitcoin Gold team said, which is why the attacker went after large, fast deposits.
Ethereum Classic, the chain that kept the original Ethereum history after the 2016 split, was hit at least twice in August 2020. In the first attack, Bitquery’s analysis found that the attacker bought hash power on the rental market NiceHash for about 17.5 BTC (some $192,000), mined in private from 16:36 UTC on 31 July and double-spent 807,260 ETC, about $5.6 million. Days later came a second one: The Block reported on 6 August 2020 a reorganisation of more than 4,000 blocks at block 10,935,622, after which Binance and Bitfly, the company behind the Ethermine pool, halted ETC deposits, withdrawals and payouts.
The arithmetic explains why this happens to small chains and not large ones. A few hundred thousand dollars of rented power can be enough to overwhelm a network whose miners are few. On Bitcoin, an attacker would have to outpace an entire industry of machines built for that one task, and on Ethereum an attacker would have to buy and lock up a large share of all staked ether, which can then be cut. Exchanges protect themselves on small chains by waiting for more confirmations before crediting a deposit.
Where crypto hacks really happen: contracts, bridges, exchanges, wallets
Layers three and four are where the large sums go. Here the blockchain records the theft faithfully, because from its point of view each transaction was properly signed.
Smart contracts and DeFi code
A smart contract is a program stored on the chain that holds and moves funds by itself. If its logic has a hole, anyone can call it in a way its authors did not foresee, and the chain will execute the result. The first famous case was The DAO on Ethereum in 2016: an insecure contract was drained of over 3.6 million ETH, according to ethereum.org’s history page. It remains the one time Ethereum changed its own history to undo a theft. A fork on 20 July 2016, at block 1,920,000, moved the funds to a recovery contract, and those who refused the fork kept running the old chain as Ethereum Classic.
Code bugs are among the most frequent incidents but rarely the costliest. CertiK counted 204 code-vulnerability incidents in the first half of 2026, more than any other category, for $151.6 million of losses, a small share of the half-year total. A recent example on Coindaro: NEAR Intents, a cross-chain swap system, lost about $3.8 million between 29 September and 1 October 2026 to a flaw in its own contract, and got all of it back on 2 October.
Bridges between blockchains
A bridge locks coins on one chain and issues matching tokens on another. The locked pool can be huge, and it is often guarded by a small group of keys. On 23 March 2022, North Korea’s Lazarus Group took almost $620 million from a blockchain project linked to the game Axie Infinity, the US Treasury said on 6 May 2022 when it sanctioned a mixer used to launder part of the money. The target was the Ronin bridge. Treasury called it the largest virtual currency heist to date. Whatever the entry point, the money left through the bridge, not through a flaw in Ethereum, where the stolen ether sat.
Exchanges and custodians
A centralised exchange holds its customers’ coins in wallets it controls. Break into the exchange’s systems and you can move those coins with valid signatures. Mt. Gox, a Tokyo bitcoin exchange, filed for protection in February 2014 after about 850,000 BTC were reported lost to hacks and theft. Its trustee moved the repayment deadline again in a notice dated 16 October 2026, to 31 October 2027.
The Bitget theft of 24 September 2026 shows how this layer fails today. Bitget says its private keys were not compromised and its cold storage was untouched. Instead, the attacker used a zero-day flaw in a third-party security product to reach internal systems, took high-privilege credentials and injected forged withdrawal commands into the wallet backend, where they passed the risk checks as normal withdrawals. Bitget covered customers from its own protection fund and restored all withdrawals on 2 October.
Wallet software, devices and signing screens
Your wallet app is software, and software can be poisoned. On 21 October 2026 the Polkadot Community Foundation warned that four versions of the Fearless Wallet browser extension, in use since February 2025, contained code that sent users’ private keys to an attacker’s server. Hardware wallets move the keys off the computer, but they still depend on companion apps, updates and the screen you approve a transaction on. The largest crypto theft on record, as NCC Group describes it, went through exactly that kind of screen.
The Bybit hack: Ethereum worked, the signing screen did not
On or about 21 February 2025, Bybit lost about $1.5 billion in virtual assets. The FBI said on 26 February 2025 that North Korea was responsible, and it calls this group of actors TraderTraitor. Trackers give slightly different totals depending on how they value the tokens, from about $1.45 billion at CertiK to $1.5 billion at Chainalysis.
The mechanism, as reconstructed by NCC Group in March 2025, went like this. Bybit kept a large reserve, including 401,347 ETH, in a cold wallet that needed several staff members to approve each transaction through Safe{Wallet}, a multisignature service. The attackers had altered the JavaScript of the Safe{Wallet} web interface. When Bybit’s signers went to approve what looked like a routine transfer from cold to hot wallet, the code swapped in a different transaction behind the screen. That transaction handed control of the wallet contract to code the attackers had written, and they emptied it.
Every signature was valid. Ethereum checked them, included the transaction in a block and carried on producing blocks as usual. Nothing in the protocol was broken. What failed was the chain of trust between a web page and the people approving a payment they could not fully read, a practice known as blind signing.
The money then went through the laundering pattern Chainalysis describes for North Korean thefts: swaps into other assets, bridges to other chains, mixers, over a cycle of about 45 days. The FBI asked exchanges, bridges and node operators to block transactions linked to the addresses it published.
Users: phishing, stolen seed phrases and physical coercion
For an ordinary holder, the most likely attacker is not a state hacking unit going after an exchange. It is someone trying to get you to hand over your seed phrase, or to sign one transaction you should not sign.
Chainalysis counted 158,000 thefts from personal wallets in 2025, affecting at least 80,000 victims, for about $713 million in its lower-bound estimate. By value, personal wallets were 44% of all crypto stolen in 2024 and about 20% in 2025, a share Chainalysis says would have been 37% without the single Bybit theft. CertiK, which uses different categories, found phishing to be the most frequent attack in 2025, with 248 incidents and $722.9 million lost.
Two incidents from the last fortnight show the routes. On 17 and 18 October 2026, an attacker emptied 1,404 Qubic wallets in under two hours because the software that created their recovery seeds had made them guessable. The same weekend, a fake security alert pushed through the LIQUID app of hardware wallet maker Ngrave sent users to a copy of its website, where those who typed in their recovery phrase lost at least $2.56 million from about 70 addresses. Ngrave says neither its app code nor its devices were breached.
The rule that protects against the second kind is simple and absolute: a seed phrase is typed only into the wallet that created it, or into a device you are restoring, never into a website or a support chat. Our guide to seed phrase security covers how to store one, and the lesson on private and public keys explains why whoever has the phrase owns the coins.
Then there is the attack no software can stop. Chainalysis estimated on 6 August 2026 that more than $30 million had been taken in violent “wrench attacks” on crypto holders in the first half of 2026, after a record $58 million in 2025. France had logged 30 publicly known incidents by mid-year, the most of any country, a surge Chainalysis puts down most likely to a data breach: in 2024 a French tax official is alleged to have stolen and sold files on wealthy crypto holders, including their addresses.
Crypto hack figures by year, and why the trackers disagree
Several firms publish theft totals. They rarely match, because each one draws the line differently: some count only hacks of platforms, others add phishing of individuals; some stop counting in early December; stolen tokens are valued at different dates; frozen or returned funds are subtracted by some and not others.
| Tracker (published) | Period | Stolen | Detail |
|---|---|---|---|
| Chainalysis (19 Dec 2024) | 2024 | $2.2 billion | 303 incidents; North Korea about $1.34 billion (61%); private key compromises 43.8% of value |
| CertiK Hack3d (2025 report) | 2024 | $2.45 billion | Baseline given in CertiK’s 2025 report |
| Chainalysis (18 Dec 2025) | January to early December 2025 | over $3.4 billion | Bybit $1.5 billion; North Korea at least $2.02 billion, a record |
| CertiK Hack3d (2025 report) | 2025 | $3.35 billion | Supply chain $1.45 billion in 2 incidents; phishing $722.9 million in 248 |
| Immunefi (Q1 2025 report) | January to March 2025 | $1.64 billion | 40 incidents, all hacks; $348 million in Q1 2024 |
| CertiK Hack3d (6 Jul 2026) | January to June 2026 | $1.32 billion | 344 incidents; about $1.2 billion net of frozen or returned funds |
| Hacken, via Cointelegraph (15 Oct 2026) | July to September 2026 | about $1.18 billion | 107 incidents; worst quarter since Q1 2025 |
One theft can make a year: Bybit alone was more than four tenths of 2025 in both Chainalysis’s and CertiK’s counts. And the biggest losses keep coming from attacks on keys and signing processes, not from the chains. In the first half of 2026, CertiK put wallet compromises first at $444.5 million in 33 incidents and phishing second at $366.3 million in 63. Its two largest cases, the Kelp DAO compromise (about $291 million) and the Drift Protocol breach (about $285 million), both in April, made up nearly 44% of the half. Our report on the third-quarter figures compares Hacken, CertiK and DefiLlama line by line.
Keep fraud apart from these numbers. Investment scams, fake platforms and romance schemes are counted separately by law enforcement and are larger still; they are covered in the scams guide rather than here.
What this means for where you keep your coins
So can a blockchain be hacked? At the base layer of a large network, rarely, and the useful question becomes: which of the other layers am I trusting? The answer depends on where your coins sit. This is not a recommendation to hold or sell anything, only a map of the risks.
- Coins on an exchange are protected by the exchange’s own security and finances. If it is hacked or fails, what you get back depends on its reserves, its insurance or protection fund, and the courts.
- Coins in a self-custody wallet depend on the wallet software, the device it runs on and on you keeping the seed phrase secret. Our guide to crypto wallets compares the options.
- A hardware wallet keeps the private key off your computer and phone, which closes the most common software attacks; our step-by-step hardware wallet guide shows how to set one up.
- Coins deposited in a DeFi protocol or sent across a bridge add the risk of that code and of whoever controls its admin keys.
- Whatever the wallet, read what you sign. If the device or app cannot show you clearly where funds are going, treat the request as suspect.
US regulators frame it the same way. In an investor bulletin of 12 December 2025, the SEC’s Office of Investor Education and Assistance warned that with third-party custody, “if the third-party custodian is hacked, shuts down, or goes bankrupt, you may lose access to your crypto assets”. In the UK, the FCA’s crypto basics page, last updated on 29 January 2026, lists cyberattacks among the reasons investors can lose everything and says crypto losses are highly unlikely to be covered by the Financial Services Compensation Scheme. In the United States, the FBI’s Internet Crime Complaint Center (IC3) is where thefts are reported; how a theft is treated for tax purposes is the subject of a later lesson on crypto taxes.
The biggest crypto hacks in history, from Mt. Gox to Bybit, get a case-by-case reading in the next lesson of this path.
