---
title: '2026 DeFi hacks: a forensic report'
description: 'Twenty-five DeFi incidents from the first half of 2026, from Drift and
  KelpDAO down to TMXTribe: how each was done, where the money went and what users
  got back. Keys, not code, caused most of the loss. Updated to November with what
  has happened since.'
slug: 2026-defi-hacks-forensic-report
date: 2026-11-18
updated: 2026-11-18
author: Stefanos Kissas
status: approved
tags:
- hacks
- forensics
- '2026'
- bridges
- oracles
- cover
facts_as_of: 2026-06-01
data: null
---

![Does anyone get their money back from a DeFi hack?](/blog/images/2026-defi-hacks-forensic-report/0213fd90-incidents.png#full "Incident resolution and user reimbursement")

N.B. This article is from the original forensic report released in June 2026 with updated data. The origninal can be found here: https://medium.com/@caducus/2026-defi-hacks-forensic-research-report-a70a2a616f1c

**In short**

- In the first half of 2026, 25 DeFi incidents took about $880 million. April was the worst month DeFi has had, with Drift ($285M) and KelpDAO ($290M) inside three weeks.
- Keys, not code, caused most of the loss: phished admins, keys stored together on one laptop, single-key mint functions and bridge verifiers that were never fit for the value they held.
- North Korea's share of crypto hack losses reached 76% through April, by TRM Labs' count. Since June the year has got worse: Binance lost $6.36 billion on 29 October.

## What defined the first half of 2026?

Not a new kind of smart-contract bug. The year was defined by human and operational failure: compromised private keys, phished admin roles, multisig keys kept on one device, and bridge verifier settings that a single compromised node could fool. Code was rarely the weakest link. People, infrastructure and process were.

Six patterns stood out.

1. **Keys, not code.** Private-key compromise and admin phishing overtook contract bugs as the leading cause of loss, by count and by value. Drift, KelpDAO, Humanity, Step Finance, Resolv, Wasabi, Gravity Bridge and Volo all trace back to operational security, not Solidity.
2. **North Korea accelerated.** TRM Labs puts its share of crypto hack losses at under 10% in 2020, 22% in 2022, 37% in 2023, 39% in 2024, 64% in 2025 and 76% through April 2026, about $577 million. The methods are slower and more precise: months of infiltration, spear phishing that impersonates real counterparties, and drains across several chains timed to beat a human response.
3. **Bridges are still the biggest single losses.** KelpDAO, Syscoin, Hyperbridge, Gravity Bridge and Saga were all bridge or cross-chain verification failures. Across DefiLlama's whole record, bridges account for about 18% of everything lost.
4. **Oracles failed in thin markets.** YieldBlox, Venus, Makina and Rhea were exploited through prices from shallow markets. An oracle that reports a shallow market correctly is itself an attack surface. The fix is a minimum-liquidity threshold before a price is trusted, not a better oracle.
5. **Old contracts became targets.** Truebit (a 2021 contract never audited) and DxSale (2021 liquidity lockers) show that aged code is being found and drained. Researchers point to AI-assisted scanning, which can check thousands of old contracts for a missing modifier or an overflow at once.
6. **DeFi rescued itself.** After KelpDAO, Aave's "DeFi United" coalition raised over $300 million (132,650 ETH) from Consensys, Mantle, the Aave DAO, Lido, EtherFi and dozens of others to restore rsETH's backing. It was DeFi's first organised cross-protocol recapitalisation.

## What were the largest incidents?

| Protocol | Date | Loss | How | Users got back |
|---|---|---|---|---|
| KelpDAO | 18 April | $290M | Two LayerZero-hosted RPC nodes behind a one-of-one verifier compromised; 116,500 rsETH minted against nothing | Indirectly, through DeFi United; about $71M frozen |
| Drift | 1 April | $285M | Months of social engineering; the timelock removed; council members tricked into pre-signing admin handover; a worthless token whitelisted as collateral | Recovery tokens, in progress |
| Humanity Protocol | 8 June | $36.4M | A phishing email infected a director's laptop holding several multisig keys | Full, through a token reissue |
| Step Finance | 31 January | $27.3M | Executive devices breached, treasury keys stolen | About $4.7M recovered; the protocol wound down |
| Truebit | 8 January | $26.2M | Integer overflow in an unaudited 2021 contract let TRU be minted for nothing | None; the protocol is dead |
| Resolv | 22 March | $25M | A key in AWS KMS let one signer mint 80M USR against $100K; pausing needed four of four | Tiered repayment in USDC |
| Rhea Finance | 16 April | $18.4M | A double-counting slippage flaw plus eight fake pools | About $9M returned or frozen |
| YieldBlox | 22 February | $10.97M | A $5 trade in a dead market lifted a collateral price 100 times | About $7.2M frozen by Stellar validators |
| THORChain | 15 May | $10.7M | A rogue node rebuilt a vault key through a flaw in its threshold signing | Absorbed by the protocol; no user loss |
| Syscoin | 7 June | $8.56M | A malformed SPV proof minted 5 billion SYS | Returned in full |

Fifteen more incidents, from $7.3 million (DxSale) down to $1.4 million (TMXTribe), followed the same lines: deployer keys with no multisig and no timelock (Wasabi), public functions with no permission check (TrustedVolumes, New Market Trading), a documented donation attack that an audit had flagged and the team had dismissed (Venus), and probable rug pulls (TesseraDAO).

Aave's CAPO glitch on 10 March, $27.78 million of wrongful liquidations from a stale oracle parameter, is not counted: no attacker took anything.

## What do the incidents have in common?

**The function that destroys value had fewer controls than the function that protects it.** At Resolv, one key could mint without limit while pausing took four. At Wasabi, one deployer key held the only admin role with a zero-second timelock. At Drift, the timelock was removed five days before the attack.

**Keys were kept together.** Humanity's multisig keys were backed up on the same laptop. A multisig whose keys share a device is a single key.

**Audits did not cover what broke.** Resolv had 18 audits and a $500,000 bug bounty. Venus had been warned about the exact attack in a 2023 audit. New Market Trading shipped the vulnerable module in an upgrade that was never re-audited.

**Old money sat unwatched.** Gravity Bridge still held millions in a bridge few people used. DxSale's lockers held more than 1,400 positions from 2021.

## What has happened since June?

The second half has been worse, and most of it has not been DeFi. On DefiLlama's record to 2 November, 2026 already counts 329 incidents and $8.78 billion: more incidents than any year before it.

| Victim | Date | Loss | How |
|---|---|---|---|
| Binance | 29 October | $6.36B | A DevOps engineer's compromised laptop gave access to a service container |
| Bitget | 24 September | $387M | Hot wallet key compromised |
| Liquid Network | 6 September | $320M | Unbacked cross-chain mint |
| Tectonic | 30 August | $124M | Donation attack |
| COLDCARD | 31 July | $116M | Weak key generation |
| Ledger | 9 October | $93M | Supply-chain attack |

The pattern held. The largest theft in history came through one person's laptop, not a contract.

## What does this mean for cover?

After KelpDAO, Aave had to raise over $300 million of private capital in about two weeks. It worked once. It is not a mechanism. The cover available in DeFi today covers a small fraction of the value at risk, and a committee or a token vote decides each claim after the fact.

Verdict Cover, which is in development, is built on what this year showed.

- **The rating prices the risk first.** Most of this year's failure modes are criteria in our ratings: a single-key admin role (Wasabi, Resolv), keys kept together (Humanity), oracle source and liquidity (YieldBlox, Venus, Makina), legacy contracts (Truebit, DxSale) and bridge verifier settings (KelpDAO). The grade sets the opening premium, from near 1% of the covered value at AAA to near 50% at D. The market prices it from there.
- **Detection does not depend on a user noticing.** Detectors flag an incident and the market pauses.
- **No single key decides a claim.** The community votes to confirm the incident happened, GenLayer validators adjudicate that confirmation, and a final community vote settles grey areas such as funds recovered in part. No committee decides, and no single key can trigger a payout.
- **No upgrade path for an attacker.** Each pool's contracts are fixed when deployed. The proxy upgrade that let Wasabi's attacker drain four chains does not exist in Cover's design.

## What should a team take from this?

- Put more controls on the dangerous function than on the safe one. Minting and upgrades need a multisig and a timelock; pausing should need less.
- Never keep two signing keys on one device.
- Re-audit every upgrade, and re-read old contracts that still hold money.
- Set a minimum liquidity before trusting any price.
- Treat a bridge's verifier setting as the bridge's security. One-of-one is one key.

## Caveats

The attributions to North Korea for KelpDAO and Humanity are preliminary, as the firms that made them say. The split of blame between KelpDAO and LayerZero is still contested. Several loss figures moved between first disclosure and final accounting; where they differ, this report keeps the figure from its June sources. TMXTribe, Moonwell and Solv rest on fewer independent sources than the larger incidents.

## Sources

- The original report on Medium: https://medium.com/@caducus/2026-defi-hacks-forensic-research-report-a70a2a616f1c (24 June 2026)
- DefiLlama hacks record: https://defillama.com/hacks (read 4 November 2026)
- TRM Labs on Binance: https://www.trmlabs.com/resources/blog/north-korea-steals-usd-6-36-billion-from-binance-in-the-largest-hack-in-history
- Rekt News: https://rekt.news
- Verdict Cover: https://www.verdict.finance/products/cover