cryptonist
Fibre optic strands gathered into a bundle, their ends lit

Research

Crypto hacks in 2026 so far: what DefiLlama's record shows

As of 22 November 2026, DefiLlama’s hacks record lists 365 incidents dated this year, with $8.8B recorded as lost: more incidents than any full year in the record, and more value. One row, the theft from Binance on 29 October, is $6.4B of that, or 72.1%. Without it 2026 is still the busiest year on record by count, but only the fifth largest by value of the nine years since 2018.

What is counted, and how

Every figure here comes from the snapshot of DefiLlama’s record that this site committed on 22 November 2026, run through a script that groups incidents by UTC date, class, technique and chain the way the hacks data pages do and takes each amount as the record gives it. The latest incident in the snapshot is dated 21 November, and the 2026 data page lists the 200 largest of the year’s rows.

Three rules shape the numbers. A class or technique of Unknown means the record has none, so it is not counted as one: a single 2026 incident has no class (Locus Finance, $18K) and 24 have no technique ($44.0M between them). An incident on several chains counts in full on each, so chain totals overlap. And DefiLlama’s hacks dashboard and its documentation give no method for the record: nothing on how an incident is added, which prices value it, or whether an amount is net of funds later returned. How crypto gets hacked sets out those limits against the whole record; this piece takes one year of it.

A year in progress has a further limit: rows arrive late. When this site last counted, on the snapshot of 3 October, the record held 286 incidents dated from 1 January to 1 October 2026. The same period now holds 292. Whatever 2026’s count is today, it is a count of what has been entered so far.

2026 against every earlier year

YearIncidentsValueLargest incidentLargest as share of year
201831$1.1BCoincheck, $534.0M47.5%
201926$277.3MUpbit, $49.3M17.8%
202046$4.0BLuBian, $3.5B88.0%
2021127$2.9BPoly Network, $611.0M21.3%
2022185$3.6BRonin Bridge, $624.0M17.2%
2023197$1.6BMixin Network, $200.0M12.3%
2024217$1.7BDMM Bitcoin, $305.0M18.3%
2025148$2.7BBybit, $1.4B51.6%
2026 to 21 Nov365$8.8BBinance CEX, $6.4B72.1%
2026 without Binance CEX364$2.5BBitget, $387.0M15.7%

On count, 2026 is not close. Its 365 incidents are 68% more than 2024’s 217, the previous high, and with about five weeks left the year already holds 26.6% of every incident in the record since 2011. Measured over the same stretch of the calendar, 1 January to 22 November, 2025 had 130 incidents and $2.6B.

On value the picture depends on one row. With Binance CEX, 2026 is more than twice 2020, the previous largest year. Without it, $2.5B puts 2026 below 2020, 2022, 2021 and 2025 and above the four other years since 2018. The record’s 31 incidents from before 2018, $931.8M in all, are spread over 2011 to 2017, and none of those years has more than eight.

Most of the count is small. Of the year’s incidents, 240 (65.8%) are under $1M, 31 of them recorded at zero, and the median incident is $400K, against $811K in 2025. At the other end, seven incidents of $100M or more hold $7.9B, 89.5% of the year’s value. Whether a rising count of small incidents means more small attacks or better recording of them is a question the record cannot answer, and the whole-record analysis treats it at length.

By class of failure

ClassIncidentsValueShare of value
Token & Share Accounting61$257.7M2.9%
Protocol Logic60$72.0M0.8%
Access Control56$371.3M4.2%
Key Compromise42$727.9M8.3%
Oracle Manipulation42$74.6M0.8%
Bridge & Cross-Chain37$703.3M8.0%
Frontend & Infrastructure23$6.5B74.0%
Input Validation22$20.6M0.2%
Governance11$47.6M0.5%
Social Engineering5$4.2Munder 0.1%
Market Manipulation2$7.8M0.1%
Reentrancy2$23Kunder 0.1%
Rugpull1$7.3M0.1%

Four of the five most frequent classes concern contract code; the fifth is Key Compromise. The three at the top, accounting errors, logic flaws and open permissions, together hold 177 incidents and $701.0M. By value the table is one class: Frontend & Infrastructure holds 74.0% of the year with 23 incidents, and all but $168.2M of that is the Binance row. Take the row out and the two largest classes of the remaining $2.5B are almost level: Key Compromise ($727.9M, 29.6%) and Bridge & Cross-Chain ($703.3M, 28.6%), with Access Control third ($371.3M, 15.1%).

Two shifts against 2025 stand out, and neither can be read at face value. Bridge & Cross-Chain has 37 incidents in 2026 against 3 in 2025, and 37 of the class’s 59 incidents across the whole record carry a 2026 date. Social Engineering moved the other way, from 8 incidents and $1.5B in 2025, almost all of it Bybit, to 5 incidents and $4.2M. The record gives each incident one class and says nothing about how classes are assigned, so a change in labelling would produce the same table as a change in attacks. Cross-chain bridge hacks looks at what that class holds.

By technique

Counted by incidents, the leading techniques are ones that take little in each incident: Spot Price Manipulation (32 incidents, $68.6M), Improper Access Control (31, $56.2M), Incorrect Share Accounting (23, $8.9M) and Private Key Compromised (21, $124.3M).

Counted by value the list is different from top to bottom. Key Leaked via Infrastructure has 9 incidents and $6.4B, Hot Wallet Key Compromised 7 and $434.2M, Unbacked Cross-Chain Mint 7 and $334.4M, Cross-Chain Message Spoofing 6 and $298.2M, and Proxy Upgrade Hijack a single incident, Drift Trade, at $295.0M. Two of the larger techniques take most of their value from rows whose target type is Wallet: Weak Key Generation (5 incidents, $123.3M, four of them against a Wallet, COLDCARD alone $116.0M) and Supply Chain Attack (4, $96.7M, two against a Wallet, Ledger (Supply Chain) alone $93.0M). Both are routes that how wallets fail describes from the holder’s side, as weak key generation and software that was not the software installed.

The largest incidents

DateIncidentClassTechniqueTargetAmount
29 OctBinance CEXFrontend & InfrastructureKey Leaked via InfrastructureCEX$6.4B
24 SepBitgetKey CompromiseHot Wallet Key CompromisedCEX$387.0M
6 SepLiquid NetworkBridge & Cross-ChainUnbacked Cross-Chain MintChain$320.0M
1 AprDrift TradeAccess ControlProxy Upgrade HijackDeFi Protocol$295.0M
18 AprKelpBridge & Cross-ChainCross-Chain Message SpoofingDeFi Protocol$293.0M
30 AugTectonicToken & Share AccountingDonation AttackDeFi Protocol$124.5M
31 JulCOLDCARDKey CompromiseWeak Key GenerationWallet$116.0M
9 OctLedger (Supply Chain)Frontend & InfrastructureSupply Chain AttackWallet$93.0M
24 AugNesaToken & Share AccountingArithmetic ErrorChain$50.0M

Dates are the record’s, in 2026. Tenth place is a tie at $40.0M between Mercado Bitcoin (2 November, Protocol Logic, technique not recorded) and Step Finance (31 January, Key Compromise, Private Key Compromised).

The nine rows come from five classes, so no single kind of failure carries the year’s largest losses. What they share is a target type. The record has only six incidents against a CEX in 2026, but those six hold $6.8B, 77.3% of the year’s value, while 261 incidents against a DeFi Protocol hold $1.2B, or 14.0%. Two of the six are the two largest rows of the year.

By chain

Ethereum has 126 of the year’s incidents, BSC 70, Base 39, Arbitrum 36, and Polygon and Solana 22 each. In all, the record names 87 chains for 2026. Forty-five incidents touch more than one of them, $7.2B between them, so the chain counts sum to 463 against 365 incidents; seven incidents ($96.8M) name no chain at all.

The overlap rule matters more this year than in any other. The Binance row lists seven chains (Ethereum, Bitcoin, Tron, Avalanche, Optimism, Arbitrum and Base), and each of them carries its full $6.4B. That is why Ethereum shows $7.4B for the year and each of the other six shows more than $6B. Without the row, Ethereum’s 125 incidents hold $1.0B, Tron’s 11 hold $488.2M, Solana’s 22 hold $414.1M, Arbitrum’s 35 hold $411.2M, Bitcoin’s 12 hold $181.2M and BSC’s 70 hold $151.0M. BSC has many incidents and little value in each, a mean of $2.2M. The record cannot say whether a chain’s count follows how much is deployed on it, how closely it is watched, or how weak its code is, so none of this ranks chains.

The two largest, in the companies’ accounts

Binance’s theft has an FBI statement behind it. On 31 October the FBI stated that North Korea was responsible for the theft of more than $6 billion in virtual assets from Binance on or about 29 October, named the activity TraderTraitor, and listed addresses that it, with the Secret Service and IRS Criminal Investigation, urged node operators, instant exchanges, DeFi services and other virtual asset service providers to block. Binance’s own account is given here as BleepingComputer reported it. On 30 October BleepingComputer reported that Binance had linked the theft to a DevOps engineer’s compromised laptop, and that the attackers had reached a service container and submitted unauthorised withdrawals. Binance said it had not yet found evidence of a leaked wallet private key, and that it would cover all losses from its SAFU fund and company funds. Its incident report of 3 November, as BleepingComputer summarised it, traced the start to a fake job interview: a coding assignment carried malware that the endpoint tools in use did not detect, and on 29 October the attackers used stolen credentials to deploy a malicious script and reach Binance’s wallet signing service. Binance put the loss at about $6.34B, against the record’s $6.36B, and said about $2.85B had been frozen or intercepted by 3 November.

Bitget has published its own notices. It says its reconciliation systems detected unauthorised transfers from some of its hot wallets at 19:05 UTC on 24 September. It runs three tiers of wallet; the breach reached part of the hot and warm tiers, and the cold wallets were not touched. That first notice put the loss at about $351.6M and said it fell within the coverage of Bitget’s User Protection Fund, then holding more than $464M. Bitget later revised the figure to $387.5M once assets taken on Zcash and Tron were counted. A status update from Mandiant that Bitget published says the attacker gained privileged access to two third-party security appliances on 24 September, planted a web shell on one, moved from it to Bitget’s production wallet job server and installed malicious packages there; it also records that Bitget had seen no evidence of private keys being compromised.

Set beside the record, the two accounts have something in common. The record’s technique for Binance is Key Leaked via Infrastructure, and for Bitget Hot Wallet Key Compromised. Both companies describe something else: an attacker inside the systems that request or authorise withdrawals, getting transactions signed, and neither company reported evidence that a private key was taken. One technique per row cannot hold that distinction, and the value tables above inherit whichever label the row carries. For a customer, both were thefts from the venue, which each company said its own funds would cover; what decides a customer’s position when a venue cannot absorb a loss is set out in what happens when an exchange fails.

What the 2026 record leaves open

The count is provisional. November so far holds 38 incidents, and incidents dated earlier in the year are still being added, as the rise from 286 to 292 for 1 January to 1 October shows. The record gives no valuation method and does not say whether an amount is net of anything frozen or returned, so each figure here is the record’s and no more than that.

The shape so far is plain enough: a great many small incidents, most of them against DeFi protocols, and a handful of very large ones, the two largest of them against exchanges. For a holder, four of the nine largest rows turn on where funds sat: on a venue, as at Binance and Bitget, or in wallets the record ties to a key-generation flaw or a supply-chain attack, as the COLDCARD and Ledger rows are labelled. That is the trade weighed in exchange or self-custody.