DeFi & Web3Guide

DeFi Risks Explained: Smart Contract, Oracle & Curator Risk

$2.79B was lost to non-exchange DeFi hacks from January 2025 to 15 November 2026, and four failures caused 82%: bridges, accounting math, admin access and stolen keys. This guide maps each risk and adds a checklist.

DeFi Risks Explained: Smart Contract, Oracle & Curator Risk
Illustration: CoinPaprika

Introduction

Oracle manipulation is the decentralized finance (DeFi) risk ↗ most guides lead with, yet it caused 4% of the $2.79B lost to non-exchange hacks between January 2025 and mid-November 2026 (DefiLlama, 2026). Bridges, accounting math, admin access and stolen keys caused 82%. Most of the money went in a few large events, with five incidents accounting for 45% of the total, and a handful of those involved no flaw in contract code at all. DeFi protocols carry no deposit insurance, so each failure lands on whoever supplied the funds, including lenders who never touched the broken asset, as Aave's lenders found on 18 April 2026. This guide measures the 2025-26 record by attack type, explains smart contract, admin key, oracle, bridge, liquidation, depeg and curator risk with dated case studies, and ends with a six-point checklist for any protocol page.

Key Takeaways

  • Four attack types, bridges, accounting math, access control and stolen keys, caused 82% of the $2.79B lost to non-exchange DeFi hacks since January 2025.
  • Oracle manipulation produced 60 incidents but only 4% of dollars, and Euler's 2023 loss was a missing health check, not a price attack.
  • Drift's roughly $285M (Chainalysis and TRM Labs count; DefiLlama lists $295M) and BonkDAO's $20M losses needed no code bug, only a few signers or a 1% vote, and no timelock.
  • One forged message on Kelp's single-verifier route released 116,500 rsETH on 18 April 2026 and put Aave's shared pool at risk.
  • Six public checks, covering audits, total value locked (TVL) history, admin keys, oracles, collateral and withdrawable cash, map to every failure point above.

What Did DeFi Hacks Cost in 2025 and 2026, and Which Attack Types Caused the Losses?

DefiLlama's hacks dataset lists 477 non-exchange incidents and $2.79B in losses between 1 January 2025 and 15 November 2026 (as of November 2026). Four attack types caused 82% of those dollars: bridge failures, accounting math, access control and key compromise. Oracle manipulation, the risk most guides lead with, caused 4% (DefiLlama, 2026).

How the Dataset Was Cut

The figures come from DefiLlama's hacks endpoint, pulled on 16 November 2026, whose newest entry is dated 14 November. Of 495 entries since 1 January 2025, 18 target centralized exchanges (CEXs) and are counted separately. The other 477 cover decentralized finance (DeFi) protocols (367), chains, wallets and tokens, and 448 carry a loss amount. DefiLlama assigns one class per incident, which forces judgement calls: it files Drift under access control although a fake collateral token also played a part. The count overstates any trend, because the dataset holds 290 entries for January to September 2026 against 111 for the same months of 2025, and its coverage widened.

Which Attack Types Cost the Most

Bridge and cross-chain failures top the ranking at $717M across 38 incidents, ahead of accounting errors ($656M, 80 incidents), access control ($502M, 85) and key compromise ($430M, 55). The five largest incidents account for 45% of all dollars, at DefiLlama's figures: Liquid Network at $320M, Drift at $295M, Kelp at $293M, Cetus at $223M and Balancer at $128M (DefiLlama, 2026). Other counters differ. Chainalysis, TRM Labs and Drift put Drift's drained funds at about $285M, LayerZero says $290M for Kelp and Chainalysis $292M, and Balancer's post-mortem counts $94.8M against $128M or more from Check Point and DefiLlama. Frequency tells a different story. Oracle manipulation ranks fourth by count with 60 incidents but totals $108M, and the median listed incident cost $0.54M. Frontend and infrastructure attacks jumped to $175M after a $93M Ledger supply-chain attack on 9 October 2026. Most hacks are small, and a few set the total. The largest 2026 cases are catalogued in the list of 2026 DeFi exploits and attack vectors ↗.

Exchange Hacks Are a Different Category

Centralized exchange hacks dwarf the DeFi record: 18 incidents total $8.71B. Binance leads, with a theft on 29 October 2026 that DefiLlama lists at $6.36B and Binance put at about $6.3B, a loss it pledged to cover from company funds (Al Jazeera, 2026). Bybit's theft on 21 February 2025, which DefiLlama lists at $1.4B and the FBI puts at about $1.5B, comes second (FBI, 2025). Bybit's forensic investigators traced it to malicious JavaScript served from Safe{Wallet}'s cloud storage, not to a smart contract (Sygnia, 2025). Bitget lost $388M from its hot and warm wallets on 24 September 2026, by its own final figure (Bitget, 2026). These are custody failures at companies, not protocol risk, but the cause repeats: stolen signing authority. Stream Finance's $93M loss in November 2025 sits in neither group, because an external fund manager lost the money (CoinDesk, 2025).

Attack typeIncidentsListed loss ($M)ShareLargest entry (DefiLlama)
Bridge & cross-chain3871725.7%Liquid Network, $320M
Token & share accounting8065623.5%Cetus, $223M
Access control8550218.0%Drift, $295M
Key compromise5543015.4%COLDCARD, $116M
Frontend & infrastructure241756.3%Ledger supply chain, $93M
Oracle manipulation601083.9%Rhea Lend, $18M
Protocol logic72541.9%Abracadabra Spell, $13M
Governance12481.7%BonkDAO, $20M
All other types511053.8%GMX V1, $42M
Centralized exchanges (separate)188,707n/aBinance, $6,360M

Data current as of November 2026.

Stat cards: $2.79B lost across 477 incidents, 82% from four attack types, $8.71B in exchange hacks, 4% from oracles
Bar chart of losses by attack type: bridge $717M, accounting $656M, access control $502M, key compromise $430M

Nine failure points sit behind these numbers, and each fails in a different way.

Which DeFi Risks Exist, and What Does Each One Break?

DeFi risk sorts into nine failure points, from the code a protocol runs to the choices of whoever curates a vault ↗. Each can empty a position without the others failing. The map below pairs every point with a 2025-26 example and a check, and the sections after it cover the points where the dollars concentrate.

Risks in the Code and the Keys

Smart contract risk is a flaw in code that holds funds. The Open Worldwide Application Security Project (OWASP) Smart Contract Top 10 for 2026 ranks access control first, business logic second and price oracle manipulation third (OWASP, 2026). Admin and upgrade risk is control, not code: whoever holds an upgrade key can rewrite a contract after users deposit. Governance risk means a vote can move a treasury. Bridge risk applies to every token that crosses chains, because the destination token is only as sound as the message that created it. The labels match the dataset above, so each risk can be sized against its row in the first table.

Risks in the Market and in Choices

Liquidation risk punishes a borrower when collateral falls. Depeg risk hits whoever holds a token that stops tracking its peg. Curator risk comes from the person who chooses which assets a vault lends against. Credit-style protocols that lend against off-chain assets add issuer default, a risk examined in how credit risk works in tokenized real-world asset lending ↗. Liquidity risk completes the list: a vault can report large total assets while little is withdrawable, a gap measured in the final section. Pools whose creators pull liquidity form a separate case, covered in how liquidity drains on DEX pools are detected ↗.

RiskWhat fails2025-26 exampleCheck before depositing
Smart contract bugMath or logic flaw drains a poolBalancer, 3 Nov 2025: $94.8M (post-mortem; Check Point $128.64M)Audit scope, code age, time in production
Admin key or upgradeA few signers control fundsDrift, 1 Apr 2026: $285M (Chainalysis, TRM; DefiLlama $295M)Signer threshold, timelock length
GovernanceA cheap vote moves a treasuryBonkDAO, 6 Jul 2026: $20MQuorum size, execution delay
OracleA wrong price is accepted as trueStream's xUSD feed, 4 Nov 2025Price source, pool depth, hardcoded values
Bridge or messagingA forged cross-chain messageKelp rsETH, 18 Apr 2026: 116,500 rsETHNumber of independent verifiers
LiquidationA price drop forces collateral sales10 Oct 2025 cascade: $19B erasedHealth factor buffer, collateral quality
DepegA token stops tracking its pegxUSD down 77% in a day, 4 Nov 2025Redemption path, pool depth
Curator or vaultA manager picks risky collateralStream-linked vault exposure, Nov 2025Collateral list, caps, timelock
Withdrawal liquidityDeposits cannot be redeemed at parMorpho vault: $598.2M assets, $0 withdrawableWithdrawable cash, utilization

Data current as of November 2026.

The first failure point is the contract code itself.

How Do Smart Contract Bugs Reach Depositors, and What Does an Audit Cover?

Smart contract risk is the chance that code alone loses user funds, with no stolen key involved. A prominent 2025 case is Balancer, where a rounding flaw in code running since 2021 drained pools on 3 November 2025. It also marks the limit of audits: they certify a named scope on a named date.

Math and Logic Bugs in Audited Code

Balancer's post-mortem puts the theft at $94.8M of user funds, while Check Point counted $128.64M and DefiLlama lists $128M, because each source counted differently (Balancer, Check Point, 2025). The cause was a rounding-direction error in the exact-output swap path of Balancer V2's Composable Stable Pools. It became exploitable only with two further conditions: rate providers and a low-liquidity pool state, which the attacker created by swapping out the pool's own token (Balancer, 2025). Balancer V3, built on a different architecture, was unaffected. OWASP's 2026 list names arithmetic errors and business logic as separate risk types.

What an Audit Does and Does Not Cover

An audit reviews specific contracts at a specific commit. OpenZeppelin, which audited earlier Balancer V2 code, stated that the vulnerable pool types were added after its audit scope (OpenZeppelin, 2025). Certora had formally verified properties of Balancer's stable pools in 2022, and its own review concluded those properties did not rule out rounding bias across repeated swaps (Certora, 2025). Drift was audited by Trail of Bits in 2022 and reviewed again in February 2026, and neither review covered the governance setup that failed (Credshields, 2026). An audit's date, scope and code version carry more information than the word "audited."

Contract code is one place to lose control; the keys that can change it are another.

Who Holds the Admin Keys, and How Long Can They Act Before Depositors React?

Control risk is a hidden exposure in DeFi: whoever can upgrade contracts, change parameters or move treasury funds can do what no bug can. Drift lost about $285M on 1 April 2026 (Chainalysis and TRM Labs count; DefiLlama lists $295M) and BonkDAO lost $20M on 6 July 2026 with no contract flaw, because a few signers or a cheap vote held that power with no delay.

Multisig and Key Control: Drift, 1 April 2026

Drift's Security Council, a multisig wallet, moved to a two-of-five signature threshold with no timelock on 27 March 2026. Attackers who had posed as a trading firm for months had already induced signers to pre-sign transactions using Solana's durable nonces, which keep a signature valid until the nonce is advanced (Chainalysis, TRM Labs, 2026). On 1 April two such transactions handed admin control to the attacker, who listed a worthless token as collateral and made 31 withdrawals in about 12 minutes, about $285M in total by TRM Labs' and BlockSec's count (TRM Labs, BlockSec, 2026). Jupiter liquidity pool (JLP) tokens worth $159.3M were the largest asset taken (Chainalysis, 2026). Drift's relaunch plan enforces timelocks and disables durable nonces (Drift, 2026).

Upgradeable Contracts and Timelocks

An upgradeable contract lets its admin replace the logic behind the same address, which OWASP lists as a risk class of its own, proxy and upgradeability vulnerabilities (OWASP, 2026). A timelock is the depositor's protection: a delay between a proposed change and its execution, during which funds can leave. Of 94 listed Morpho Vaults V1 on 16 November 2026, 71 use a three-day timelock, 15 use seven days and the other eight use four to 14 days (Morpho API, 2026). Fee changes on those vaults take effect immediately, so the delay covers risk changes and not every change.

Governance Attacks: BonkDAO, 6 July 2026

BonkDAO's treasury moved on a legitimate vote. An attacker spent about $4.4M buying just over 1% of BONK supply, enough to meet quorum, and the proposal passed with seven wallets voting, a 2.9% turnout. It executed automatically and moved about $20M (CoinDesk, 2026). The decentralized autonomous organization (DAO) had no execution timelock and no emergency veto (ETHNews, 2026). DefiLlama lists 12 governance incidents worth $48M since January 2025, small beside key and bridge failures. Beanstalk's $182.2M loss in April 2022 used borrowed votes to pass a proposal the same way (CertiK, 2023).

Keys decide who can change the rules; price feeds decide what those rules believe.

How Does Oracle Risk Turn a Price Feed Into a Loss?

Oracle risk is the chance that a protocol accepts a wrong price as true. OWASP ranks price oracle manipulation third in its 2026 list (OWASP, 2026), and DefiLlama counts 60 such incidents since January 2025, 46 of them spot-price manipulation. They cost $108M in total, an average of $1.8M each (DefiLlama, 2026).

Manipulable and Hardcoded Prices

A protocol that reads its price from one thin pool lets an attacker move that price and then borrow ↗ against it. DefiLlama's largest 2026 entries of this kind are Rhea Lend ($18.4M, April), Bonzo Lend ($9.1M, July) and Moonwell ($8.7M, August) (DefiLlama, 2026). Drift's attackers manufactured a price history for a worthless token by seeding a pool with roughly $500 of liquidity and wash-trading it near $1 for weeks (TRM Labs, 2026). Silo's report on Stream Finance found that xUSD feeds still showed pre-loss values after the collapse, so liquidations never fired (Silo, 2025). Morpho's documentation names the general failure: an oracle that overestimates collateral loses depositors' assets (Morpho, 2026). OWASP's listed defences are aggregated sources, time-weighted prices, liquidity checks and circuit breakers.

What Is Not Always an Oracle Attack

Labels matter, because the fix depends on the cause. Euler lost about $197M in March 2023 to a donation function with no health check, not to a manipulated price (Omniscia, 2023). On 17 September 2026 an attacker inflated the price of NSTR on a thinly traded pool that Nostra's price feed read. DefiLlama classifies it as oracle manipulation, and the published account describes no flash loan, as the Nostra fake-pool incident ↗ explains. Counts by class therefore carry judgement. Drift sits under access control although a fake price completed the theft.

Bar chart of incident counts: access control 85, accounting 80, logic 72, oracle manipulation 60, key compromise 55

A bridge is an oracle for another chain's state, and it fails the same way.

How Do Bridge and Cross-Chain Messaging Risks Reach Other Protocols?

Bridge and cross-chain failures are the largest dollar class in the dataset: 38 incidents and $717M since January 2025 (as of November 2026). A bridged token is a claim on a message. When the message is forged, the claim is worthless yet still sits in lending markets as collateral.

Kelp's Single-Verifier Route

On 18 April 2026 an attacker forged a LayerZero message, and Kelp DAO's Ethereum adapter released 116,500 rsETH, about $292M (Chainalysis, 2026). The rsETH route required one verifier, the LayerZero Labs decentralized verifier network (DVN). LayerZero says attackers compromised two remote procedure call (RPC) nodes that the DVN read from and knocked out an external one with a distributed denial-of-service (DDoS) attack, so the DVN confirmed a burn that never happened (LayerZero, 2026). A second forged packet for 40,000 rsETH was blocked after Kelp paused its contracts (Blockaid, 2026). Responsibility is disputed: LayerZero says it advised multiple verifiers, while Kelp says one verifier was the documented default (OAK Research, 2026).

How rsETH Reached Aave

Of the stolen rsETH, 89,567 was posted as collateral on Aave, backing about 82,650 WETH of borrowing (Aave Governance, 2026). Aave's contracts worked as designed, and guardians froze rsETH markets in roughly 90 minutes. LlamaRisk modelled $123.7M to $230.1M of potential bad debt, and stablecoin markets reached 100% utilization (LlamaRisk, Galaxy Research, 2026). A coalition called DeFi United refilled rsETH's backing in tranches through late May (Messari, 2026). Liquidating the attacker's positions on 6 May left Aave's main Ethereum market with a 52,964 WETH deficit, about $95 million, still recorded on-chain on 16 November 2026 (Aave on-chain data, 2026). A deficit offset raised to 53,041 WETH in the same transaction keeps it from slashing stakers in Umbrella, Aave's staker-funded backstop, and the recovery plan said ETH from the recovered rsETH would clear such deficits (Aave, 28 April 2026). Lenders who never held rsETH met withdrawal limits because one collateral sat in a shared pool. The restaking layer behind such tokens has its own guide, how restaking works on EigenLayer ↗.

Liquid Network's Federated Peg

Bridge risk also covers sidechains. On 6 September 2026 a flaw in the rangeproof cache of Elements, the software behind Liquid Network, let a transaction mint about 4,000 unbacked Liquid Bitcoin (L-BTC). It was redeemed through a peg-out key held by SideSwap, and the 15-member federation released about 3,996 BTC, cutting the reserve from roughly 4,205 BTC to 197 BTC (Blockstream, 2026). Federation keys were not stolen; consensus had accepted the fake balance. The attacker returned 3,400 BTC on 7 September, leaving about 602 BTC outstanding. Chainalysis valued the withdrawal at $320M (Chainalysis, 2026).

Flowchart from poisoned RPC nodes to a forged burn, 116,500 rsETH released, Aave collateral, 82,650 WETH borrowed

Cross-chain failures mint bad collateral, and ordinary market stress breaks good collateral equally well.

How Do Liquidations, Depegs and Curator Choices Spread One Failure to Many Users?

Market and credit risks work through collateral. A price drop, a broken peg or a failed issuer reaches lenders who never held the asset, because lending markets accept it as security ↗. The three mechanisms below each have dated examples.

Liquidations and Forced Selling

A borrower who posts 10 WETH on Aave V3 and borrows 11,000 USDC is liquidated when WETH falls about 27%, from $1,798 to $1,310 (Aave PoolDataProvider, own calculation, 2026). Liquidation is permissionless, so bots compete to repay debt for a 5% bonus. At scale the forced sales push prices lower. The 10 October 2025 cascade erased $19B across crypto in 24 hours (CoinDesk, 2025), and Hyperliquid's auto-deleveraging closed about $2.1B of positions in 12 minutes (Chitra et al., 2026). The mechanics of that event are in the liquidation cascade analysis ↗.

Depegs: stETH, USDe and xUSD

stETH traded at 0.931 of ETH ↗ on Curve on 18 May 2022 and at a record 8% discount in Dune data on 13 June, with no redemption path until withdrawals opened on 12 April 2023 (arXiv, CoinDesk, Lido, 2023). Exit is better now but the market is thin: selling 20,000 stETH through Curve's pool cost 28.4% on 16 November 2026 (Ethereum mainnet contracts, 2026). Venue prices can also mislead. USDe printed $0.65 on Binance on 10 October 2025 while Ethereum pools held near $0.99 (Ethena Governance, 2025). Stream's xUSD fell 77% within a day on 4 November 2025 (Decrypt, 2025).

Curator and Vault Choices

A curated vault puts a manager between the depositor and the collateral. Stream Finance disclosed on 4 November 2025 that an external fund manager had lost about $93M, and xUSD was collateral in curated markets on Euler, Morpho and Silo (CoinDesk, 2025). Analysts at Yields and More estimated about $285M of Stream-linked debt and called the map incomplete (Decrypt, 2025). Euler Labs said the protocol operated as designed (Euler Labs, 2025). Listing an issuer-priced asset was the curator's choice, and depositors bore the loss. Vault totals mislead too: one unlisted Morpho vault showed $598.2M of assets and $0 withdrawable on 16 November 2026 (Morpho API, 2026). Defaults at off-chain borrowers follow the pattern in real-world asset default events in DeFi ↗.

Each of these failures was readable on a public page before it happened.

What Should a Depositor Check Before Putting Money Into a DeFi Protocol?

Six checks catch most of the failure points in this dataset, and each can be read from public pages in minutes: audits and age, total value locked (TVL) history, admin keys and timelocks, oracle design, collateral concentration and withdrawal liquidity. No check proves safety. Each removes one way to lose money.

Six Checks Before Depositing

The first check is the code's age and audit scope, then the TVL chart on DefiLlama: a protocol whose TVL arrived last week has no survival record. Next comes who controls upgrades and how long a timelock runs, followed by how each price is sourced and which assets back a lending market or vault, including bridged and issuer-priced ones. The last check is withdrawable liquidity, not total assets. Time in production is a filter, not a guarantee, as Balancer's years-old pool code showed. Collateral checks matter because lenders inherit the weakest listed asset, and liquidity checks matter because vault totals count claims on borrowers, not cash. The table pairs each check with a red flag and a place to read it.

Reading a Multisig and a Timelock

A multisig is a wallet that needs several signatures. Three questions apply: how many signers are required, whether they are independent people on separate devices, and what delay applies. Drift's two-of-five council with no timelock fell to two compromised signers and no waiting, and the attackers reached those signers through months of social engineering and infected laptops (Credshields, 2026). A three-day delay is the common setting among Morpho's vaults, and it gives depositors time to exit (Morpho API, 2026). A protocol that publishes signer identities, thresholds and delays has made the check possible. One that hides them has not.

What Cover and Position Size Do Not Fix

Discretionary cover, such as mutual cover pools ↗, pays claims at the pool's discretion and carries exclusions, so its terms need reading per protocol. Nexus Mutual, a discretionary mutual, received no claims for Kelp and classes the exploit as a bridge and operational-security failure outside cover; it also received none for Balancer (Nexus Mutual, 9 October 2026). Aave Umbrella, the lending protocol's own backstop, offsets bad debt automatically but covers only WETH, USDC, USDT and GHO on Aave V3's Ethereum core market, so it is not a payout guarantee for depositors (Aave Docs, 2026). Position size is the other control. Lenders to Aave's shared pool in April 2026 never held rsETH yet met frozen withdrawals, so exposure is wider than the visible deposit. Spreading money across protocols helps against single failures and not against shared collateral: receipt tokens such as weETH, wstETH, rsETH and osETH made up $5.8B of Aave V3's $14.0B TVL in mid-November 2026 (DefiLlama, 2026). Rug pulls, token approvals, maximal extractable value (MEV) and audit reports are separate subjects with their own checks.

CheckWhat to look atRed flagWhere to read it
Audits and ageAudit date, scope, code version, months in productionCode changed since the audit; deployed last monthProtocol docs, audit report, block explorer
TVL history12-month TVL line, share held by one walletSudden spike; one depositor dominatesDefiLlama protocol page
Admin keys and timelocksWho can upgrade, signer count, delayOne key; two-of-five with no delay; upgradeable proxyBlock explorer, docs, governance forum
Oracle designPrice source, pool depth, time-weightingSingle thin pool; hardcoded or issuer-set priceProtocol docs, market page
Collateral concentrationAssets behind loans; bridged or issuer-priced onesOne asset dominates; single-verifier bridgeMarket or vault page, curator disclosures
Withdrawal liquidityWithdrawable cash against total assets; utilization100% utilization; exit queueVault or market page

Data current as of November 2026.

Conclusion

DeFi risk is a map of dependencies, not a single danger. Code, keys, prices, bridges, collateral and curators each decide whether a deposit can leave, and the 2025-26 record shows losses concentrating where control sits with few parties: one verifier, one multisig, one governance quorum. Incident counts rise with dataset coverage and dollar totals follow a handful of events, so a single year's ranking says less than the pattern behind it.

The difference between a sound protocol and a fragile one is rarely visible in calm markets. Drift, Kelp and Stream Finance all worked until one party failed: two signers, one verifier, one fund manager. Depositors saw yield and total value locked, which measure activity, not resilience. Resilience shows up mainly as a delay, such as a timelock, a second verifier or an independent price, and by the time a missing delay is noticed the money has moved.

Before depositing, read who can change the contract and how long a change takes to execute, which prices each loan relies on, and how much of the balance can leave today. Size each position as if its weakest collateral will break, because lenders in Aave's pool in April 2026 lost access to funds without holding the failed token. A protocol that publishes none of this has answered the question for you.

Quick Stats

  • $2.79B — non-exchange DeFi losses in 477 incidents, 1 January 2025 to 15 November 2026
  • 82% — share of those dollars from bridge, accounting, access-control and key failures
  • 60 — oracle manipulation incidents, totalling only $108M in losses
  • $285M — drained from Drift on 1 April 2026 through pre-signed admin transactions (Chainalysis and TRM Labs count; DefiLlama lists $295M)
  • 116,500 rsETH — released by one forged message on 18 April 2026
  • 71 of 94 — Morpho Vaults V1 using a three-day timelock on 16 November 2026

Data current as of November 2026.

FAQ

?Is a DeFi hack the same thing as a rug pull?

No. A hack means an outsider exploits a flaw or steals a key, as in Balancer on 3 November 2025 or Drift on 1 April 2026. A rug pull means insiders drain a project they control. DefiLlama separates the two: its rug pull class holds six incidents and $11.6M since January 2025, against 477 non-exchange incidents in total (DefiLlama, 2026). Classification involves judgement, so some insider cases may sit under access control or key compromise.

?Do depositors get their money back after a DeFi hack?

Sometimes in part, and never by right. Liquid Network's attacker returned 3,400 of about 4,000 BTC within a day (Blockstream, 2026). Drift's recovery plan relies on a pool funded by exchange revenue and proposed contributions of up to $127.5M from Tether and $20M from other partners, not on guaranteed repayment (Drift, 2026). DeFi has no deposit insurance, so recovery depends on the protocol, its partners and law enforcement.

?Is DeFi riskier than keeping crypto on a centralized exchange?

The two fail differently. Since January 2025 DefiLlama lists 18 exchange hacks totalling $8.71B and 477 non-exchange incidents totalling $2.79B (DefiLlama, 2026). Exchange users face custody risk, as in Binance's roughly $6.3B theft in October 2026, Bybit's roughly $1.5B loss in February 2025 and Bitget's $388M in September 2026 (Bitget's final figure). DeFi users face code, key and collateral risk but hold their own assets. Bitget's user protection fund covered its loss (Bitget, 2026).

?Where can a depositor look up whether a protocol has been hacked before?

DefiLlama's hacks page lists incidents by date, protocol, chain, class and amount, and the same data is available through its public API. Protocol blogs give the cause, which dataset labels can obscure: DefiLlama classifies the Nostra incident of 17 September 2026 as oracle manipulation, while the published account describes an inflated NSTR price on a thin pool and no flash loan. Chainalysis, TRM Labs and BlockSec publish forensic write-ups. A clean record on a young protocol says little, so check its age as well.

?Does a hardware wallet protect against protocol risk?

No. A hardware wallet protects the keys that sign a user's transactions, and none of the protocol failures above involved a depositor's key. Balancer's rounding flaw, Kelp's forged message and BonkDAO's vote all moved funds held inside contracts. A hardware wallet helps with a different failure, because Bybit's theft began with signers approving a disguised transaction (Safe Ecosystem Foundation, 2025). Token approvals are a separate risk.

?Does a higher yield always mean higher DeFi risk?

Not always, but yield should be traced to its payer. Returns funded by token emissions or leverage loops carry collateral and liquidation risk, while returns from fees reflect trading or borrowing demand. Stream Finance's xUSD earned returns through an external fund manager, and a $93M loss cut its price 77% in a day (Decrypt, 2025). A high rate with no named payer is a signal to run the six checks first.

?How do people limit the damage if one protocol fails?

Position size and collateral variety matter more than the number of protocols. Splitting money across protocols helps against a single failure but not against shared collateral: receipt tokens such as weETH, wstETH, rsETH and osETH made up $5.8B of Aave V3's $14.0B in total value locked in mid-November 2026 (DefiLlama, 2026). Prefer positions that can be withdrawn on demand, avoid loops that add liquidation points, and favour protocols with timelocks.

References / Sources

Market & Platform Data

Loss datasets, vault data and news reports used for dated figures.

  • DefiLlama: Hacks dataset, 2025-26 incidents (defillama.com, Nov 2026)
  • Aave V3 Pool and Umbrella: WETH reserve deficit and deficit offset, on-chain (etherscan.io, Nov 2026)
  • Morpho: Vaults V1 timelock and vault data (morpho.org, Nov 2026)
  • CoinDesk: BONK treasury drain after malicious proposal (coindesk.com, Jul 2026)
  • Decrypt: Stream xUSD plunges after $93M loss (decrypt.co, Nov 2025)
  • FBI: North Korea responsible for Bybit hack (fbi.gov, Feb 2025)
  • Al Jazeera: Binance says $6.3bn stolen in biggest crypto heist in history (aljazeera.com, Oct 2026)
  • BleepingComputer: Bybit hacked via Safe{Wallet} developer machine (bleepingcomputer.com, Feb 2025)
Incident Reports & Post-Mortems

Primary and forensic accounts of the case studies.

  • LayerZero: KelpDAO incident statement (layerzero.network, Apr 2026)
  • Aave: rsETH recovery technical plan (x.com, Apr 2026)
  • Blockstream: Liquid Network security incident assessment (blog.liquid.net, Sep 2026)
  • Balancer: Nov 3 exploit post-mortem (medium.com, Nov 2025)
  • Drift: Incident recovery update (drift.trade, Apr 2026)
  • Chainalysis: Lessons from the Drift hack (chainalysis.com, Apr 2026)
  • Check Point Research: Balancer rounding error analysis (research.checkpoint.com, Nov 2025)
  • Nexus Mutual: public API, claims and cover data (api.nexusmutual.io, 9 Oct 2026)
  • Aave: Umbrella documentation (aave.com, Oct 2026)
Security Research & Standards

Standards bodies and audit firms on failure classes and audit scope.

  • OWASP: Smart Contract Top 10, 2026 (scs.owasp.org, 2026)
  • OpenZeppelin: Understanding the Balancer v2 exploit (openzeppelin.com, Nov 2025)
  • Certora: Breaking down the Balancer hack (certora.com, Nov 2025)
  • Credshields: Drift incident post-mortem (discover.credshields.com, Apr 2026)
  • TRM Labs: Drift Protocol heist analysis (trmlabs.com, Apr 2026)
  • BlockSec: Drift multisig governance incident (blocksec.com, Apr 2026)
Share this guide
Written by Bartek Hagan

Original CoinPaprika guide.

Report a correction
All guides

Cryptocurrencies are highly volatile and involve significant risk. You may lose part or all of your investment.

All information on Coinpaprika is provided for informational purposes only and does not constitute financial or investment advice. Always conduct your own research (DYOR) and consult a qualified financial advisor before making investment decisions.

Coinpaprika is not liable for any losses resulting from the use of this information.