Take 10,000 USDC supplied on Aave. On 1 November 2026, at about 09:45 UTC, Nexus Mutual’s public price feed quoted 110 dollars for a year of cover on that amount. If the contract is exploited, you wait out a cool-down, put down a 0.05 ETH deposit and file a claim, and three named people vote on it. That is what DeFi insurance sells today, and the most useful thing to learn about it is what happens between the exploit and the payment.
Can you insure your crypto in DeFi? Partly. You can buy cover against a few named risks, at a price that moves, in limited size, from a mutual that says in its own FAQ: “Nexus Mutual cover is not a contract of insurance.” Phishing, a stolen key and a rug pull by a protocol’s owners are all outside the standard wording. No public deposit guarantee stands behind any of it.
The size of the gap explains the rest. DefiLlama counts about 8.9 billion dollars lost by DeFi protocols from 2021 to 1 November 2026, and Nexus Mutual, the largest name in this field, says its members have paid more than 18.5 million dollars in claims since 2019, as of August 2026. That works out at about 0.2 percent. Read it as an order of magnitude rather than a coverage ratio, because the two figures have different perimeters and Nexus only answers for what it listed and sold. It still suggests that most of the money lost in DeFi had no cover behind it.
This lesson follows our walk through the major DeFi hacks and how each one ran, so it leaves the attacks aside and looks at the money afterwards: who pays, who decides and what the wording refuses. Risks that cover does not touch are in our map of DeFi risks.
How DeFi insurance works when the insurer is a mutual
A traditional policy is a contract: the insurer owes you money if the insured event happens, and a court can make it pay. Most DeFi cover is built differently, so the vocabulary matters. Buyers in this market purchase “cover”, and the word is chosen with care.
Discretion, not a legal obligation
Nexus Mutual calls itself a discretionary mutual: members blockchain that runs exactly as written when it is called, holding and moving funds according to its code, without an intermediary able to change the outcome.">smart contract holding reserves of two or more tokens, supplied by users, that a decentralized exchange uses to execute trades automatically at a price set by a formula.">pool capital and share risk, and the cover is an insurance-like product that carries a discretion to pay, not a legal obligation. The wording says the same in its verbs. Clause 7.1 of the Single Protocol Cover terms opens with “Nexus Mutual may pay the Claim Amount”, and the conditions follow. A refused buyer can refile with more evidence, but nothing in that wording promises a payment.
Three roles: buyer, staker, pool manager
The buyer picks a listing, an amount in ETH, USDC or cbBTC, and a period, then pays a fee. The cover arrives as an NFT in the buyer’s wallet, and a partial claim leaves the rest of it alive.
The other side is the staker, who deposits the mutual’s token, NXM, in a staking pool run by a pool manager. The manager chooses which listings the pool backs and at what target price. Nexus says half the cover fee goes to the stakers who underwrite that cover, and the other half accrues to the capital behind every NXM token.
Each NXM staked opens two NXM of capacity, which is why a loss is split. Half of a claim is paid by burning the stake of those who underwrote it, and half is socialised across all NXM holders through the shared capital pool. That pool holds ETH, USDC and cbBTC, and part of it is invested: members approved an allocation of 5,160 ETH to a Steakhouse vault on Morpho, according to Nexus’s review of 2025. The institution that protects you against DeFi risk therefore carries some DeFi risk of its own.
Who decides a claim, and how long it takes
Older explainers say NXM holders vote on claims. That has been out of date since November 2025. The claims documentation, last reviewed in August 2026, describes three eras. A few products name a Designated Claim Assessor instead of the committee.
| Period | Who decided | Rule |
|---|---|---|
| 2019 to March 2023 (V1) | Any member staking NXM as an assessor | 70 percent consensus, 36-hour vote |
| March 2023 to November 2025 (V2) | Members staking NXM in the assessment contract | Stake-weighted majority |
| Since November 2025 | Claims Committee of three named assessors | 2 of 3 accept votes approve |
The three named assessors are Hugh Karp, the founder, and Roxana Danila, the chief technology officer, both Advisory Board members, plus the Head of Risk, who goes by BraveNewDeFi. So the people deciding are insiders of the mutual, and their names are public. The FAQ still speaks of members having the final say, while the claims page tells you who holds the vote today.
The calendar is short but it is not instant. The standard wording imposes a 14-day cool-down after the event, during which no claim can be filed, and an annex can change that length. You then file with the 0.05 ETH deposit, about 113 dollars at that morning’s ETH price of about 2,270 dollars. An accepted claim returns the deposit; a denied claim keeps it. The vote lasts up to 72 hours and closes early once all three have voted, though governance can extend it with a fresh 72-hour window. A 24-hour pause follows, in which the Advisory Board can reverse a fraudulent vote. File on the first possible day, let the vote run its normal course, and the first possible withdrawal comes about 18 days after the event: 14 plus 3 plus 1. You then have 30 days to withdraw. A denied claim can be refiled with more evidence. At Arcadia, below, claims were filed 14 days after the exploit and Nexus announced the payout on day 20.
Discretionary or parametric: the difference in two sentences
Parametric cover pays when a measurable trigger fires, with no assessment of your actual loss, while discretionary cover has someone read the evidence and decide. Nexus sits closer to the second, even for products that look mechanical: its Crypto Cover terms contain precise triggers for a stablecoin depeg or a custody loss, yet they still say the mutual “may approve a Claim”, and the claim still goes through assessment.
Nexus Mutual in practice: KYC, prices and a record of claims
Joining means showing your ID
This is the first surprise for a DeFi user. To buy cover you become a member: identity verification with a government ID and a membership fee of 0.0020 ETH, and a single address can join only once. The restricted list of 1 November 2026 has 27 jurisdictions, among them China, India, Japan, Mexico and Russia; the United States, the United Kingdom, France and Italy are not on it. You cannot buy cover for someone else, even a spouse. The app works from Ethereum mainnet, though the cover can protect positions on other EVM chains.
What cover costs on 1 November 2026
Prices come from the public quote API behind the Nexus app. They are annual, for 10,000 USDC over 365 days, paid in USDC, and they change with capacity and demand. The table is one reading at about 09:45 UTC.
| Listing | Annual price | Premium on 10,000 USDC |
|---|---|---|
| Uniswap v3 | 0.50% | $50 |
| Sky USDS Depeg | 0.50% | $50 |
| Aave v3 | 1.10% | $110 |
| Coinbase Custody | 1.75% | $175 |
| Aave v4 | 2.09% | $209 |
| Binance Custody | 2.47% | $247 |
| EigenLayer v1 | 3.30% | $330 |
| Hyperliquid Core | 3.50% | $350 |
| Curve All Pools | 4.00% | $400 |
| Spark Lend | 4.00% | $400 |
Two details from the same minute. A 90-day quote on Aave v3 came in at 0.88 percent a year, or 21.70 dollars for the quarter, against 110 dollars for the full year, so the term changes the rate. And the API answered “not enough capacity” for 10,000 USDC on Lido, Safe and Curve’s crvUSD depeg listing. Price rises as a pool fills up, and when capacity runs out there is no price at all.
What it has paid, and what it refused
| Event | Date | Paid (dollars at payout) |
|---|---|---|
| Yearn yDAI hack | Feb 2021 | 2,317,776.72 |
| TribeDAO / Rari Fuse hack | Apr 2022 | 5,089,889.24 |
| Perpetual Protocol v1 economic failure | May 2022 | 377,147.19 |
| Hodlnaut halted withdrawals | Aug 2022 | 1,113,211.17 |
| FTX halted withdrawals | Nov 2022 | 4,924,887.90 |
| BlockFi halted withdrawals | Nov 2022 | 30,590.64 |
| Euler exploit, protocol cover | Mar-Apr 2023 | 2,389,227.88 |
| Euler exploit, Sherlock Excess Cover | 2023 | 1,000,000 |
| Arcadia Finance exploit | Jul 2025 | 236,486.21 |
| Stream Finance fallout, six claims | Filed Nov-Dec 2025 | 70,449.95 |
| Resolv exploit, Morpho vault | Filed Apr 2026 | 3,060.53 |
| Goldfinch (36,692.87 claimed) | Filed Feb 2025 | Denied |
| Beefy (5,141.76 claimed) | Filed 28 Aug 2026 | Denied |
Figures come from Nexus’s claims history and claims page, read on 1 November 2026. The two do not always match: the 24 FTX rows add up to 5.24 million dollars on the claims page, against the 4.92 million in the docs. The claims page lists 216 claims, 103 approved and 113 denied. That split needs care, because 68 of the denials were filed in 2020 and the docs warn that users tested the system in its first two years. Counting from 2023, it is 53 approved and nine denied.
FTX and Hodlnaut were custody failures, so some of the largest payments went to people who lost funds at a company and had bought cover for that, with no DeFi contract involved.
Euler shows how a payout unwinds. Nine protocol cover claims were paid for 2,389,227.88 dollars. Members who took a payout agreed to hand their eToken and dToken balances to the mutual’s treasury, and when the attacker returned funds, Nexus recovered 99.05 percent on DAI-denominated claims and 32.39 percent on ETH-denominated ones, by the count on that page. Eight of the nine members said they would return their payments. One, who redeemed 199.6 ETH, said no. If a hack is later reversed, the cover money is meant to go back.
Arcadia Finance, on Base, is the recent small case. DefiLlama puts the exploit of 15 July 2025 at 2.5 million dollars, and press reports at the time said about 3.5 million. Three claims went in on 29 July, and on 4 August Nexus said it had paid about 250,000 dollars, mostly on cover sold through OpenCover, an independent distributor. The claims page shows 236,486.21 dollars. Somewhere between 7 and 10 percent of the loss came back through cover, well above the 0.2 percent order of magnitude at the top of this page.
The claims since then show what a liquidation failure means in practice. After the Stream Finance blow-up of late 2025, the committee approved six claims for 70,449.95 dollars, most on Beefy vaults that had lent into markets where bad debt built up and a fixed-rate oracle left lenders unable to withdraw. In April 2026 it approved 3,060.53 dollars for a Morpho vault hit by the Resolv exploit, under the Liquidation Failure provision, although the debt was still unrealised; the claimant can return the money if reimbursed.
That April claim is the only one the record shows as paid in 2026, a year in which DefiLlama counts 1.19 billion dollars lost by DeFi protocols so far. No claim on the page names Kelp, Drift or Balancer.
The newest entry went the other way. Claim V2 #37, for 5,141.76 dollars, was filed on 28 August 2026 on Beefy Multi Protocol Cover bought on 23 July. The claimant had funds in two msETH/WETH Curve vaults through Beefy and lost money after MetronomeDAO, a protocol under those vaults, was drained by arbitrage on delayed oracle prices. Metronome was not in the listing’s annex, so the committee found the loss fell outside the Beefy, Curve, Convex and StakeDAO contracts and denied it.
A vault can fail through a dependency, and the annex decides which dependencies count. Credit is another gap: a Goldfinch claim of 36,692.87 dollars, filed in February 2025, was denied because the loss came from a default on an uncollateralised loan, and Protocol Cover does not cover credit risk.
How big the mutual really is
The capital pool stood at 94.4 million dollars on DefiLlama on 1 November 2026. Summing the public cover records that day gives about 75 million dollars of active cover, in 252 covers held by 87 members, an approximation from our own count that uses ETH and BTC prices of the same morning. Against about 85 billion dollars locked across DeFi, that is about 0.09 percent. Aave v3 gives a direct comparison: about 5.45 million dollars of active cover on 16.35 billion dollars of Aave TVL, or 0.03 percent.
Nexus’s business in 2025, by its own review: over 5.7 million dollars of cover fees and claims above 370,000 dollars. Claims were about 6.5 percent of fees, though that is not a loss ratio in the usual sense, because fees are shared with stakers.
Sherlock: from staking cover to audits and a small Shield
Sherlock is the second name readers meet, and many explainers still describe a product that no longer runs. Its V1 and V2 protocols are formally deprecated effective 10 September 2026. The notice says they have been inactive from a support perspective since 2024 and tells users to withdraw any remaining funds. The contracts stay deployed, and DefiLlama still shows about 230,000 dollars in them.
That model paired an audit with staked capital behind the audited code. DL News reported on 14 April 2023 that the staking pool quickly drew 30 million dollars after opening in late 2021, and that Sherlock paid out 4.5 million dollars on Euler, leaving about 2.9 million dollars on its dashboard. Co-founder Jack Sanford told the outlet: “Payouts like Euler are an existential risk to Sherlock’s survival”. On those figures, one payout took about 60 percent of what the pool held at the time.
The two names meet here. Sherlock had bought excess cover of its own from Nexus, and Nexus paid 1 million dollars on it. A first claim was denied because it ignored a 500,000 dollar deductible in the wording; a corrected claim passed.
Today Sherlock sells security work, including collaborative audits, an AI-assisted Audit Engine and bug bounties. Its docs mark audit contests as deprecated, replaced by that engine. If you want to judge its reports, the method is in our guide to reading an audit report. The remnant of cover is Sherlock Shield, offered to qualified codebases after an audit, and not with every audit. It is a payout tied to the audit’s findings, from 500,000 dollars for a clean result down to 1,000 dollars at 30 or more points, and the terms add that “Sherlock does not guarantee payment or the availability of funds”. The docs describe terms agreed with the audited team and no product a token holder can buy.
The rest of the field: quiet projects and protocol backstops
DefiLlama’s insurance category, read on 1 November 2026, puts Nexus first and Firelight second, at about 65 million dollars. Firelight launched as an XRP staking vault whose announced second phase has that stake back DeFi cover. Behind them the list thins out: Unslashed about 3.4 million dollars, Sherlock 0.23 million, InsurAce 0.14 million, Neptune Mutual zero, among others. A TVL figure tells you how much money sits in contracts and nothing about whether a team still answers claims, so a reader would need to confirm that any of them sells cover today.
The other form of DeFi insurance is built into a protocol, and the buyer is not you. Aave‘s Umbrella asks users to stake assets that can be burned automatically to offset bad debt, after a first-loss amount taken by the DAO. Its limit is in the design: each deployment protects one asset on one network. The older stkAAVE module is due to lose its slashing once Umbrella reaches sufficient scale. Those stakers are underwriters in all but name.
What cover does not cover
Anyone comparing DeFi insurance should read the wording rather than the listing title. Here is the standard Single Protocol Cover, clause by clause. It pays for a smart contract bug that lets the protocol be used in an unintended way, and for a sudden and widespread economic event caused by oracle failure, oracle manipulation, liquidation failure or a governance takeover.
Exclusions and deductible in the standard wording
| Clause | Loss from | Paid? |
|---|---|---|
| 10.1 | Non-EVM vulnerabilities, unless the annex says otherwise | No |
| 10.2 | Phishing, stolen keys, malware, or anything where the protocol keeps acting as intended | No |
| 10.4 | Events before the cover period, or after a public bug disclosure or warning | No |
| 10.5 | Market price moves, apart from oracle failure or manipulation | No |
| 10.6 | Depeg of an asset the protocol itself generates | No |
| 10.7 | Owners or controllers taking funds within their permissions (a rug pull), whoever holds their keys | No |
| 10.8 | User interface or website errors, where the protocol keeps acting as intended | No |
| 10.9 | Failure of components used to bridge assets between chains | No |
| 10.11 | Team use of admin controls, such as pausing, that leads to a loss | No |
| Definitions | First 5 percent of the cover amount (the deductible), unless the annex differs | No |
Read that table with a real loss in mind. If you sign a drainer transaction, clause 10.2 applies, and our guide to what to do after a theft is the place to go. If a team empties a pool with powers it always held, clause 10.7 applies, and that is the rug pull scenario. The bridge clause matters as well: 2026’s largest DeFi losses include one that started at a bridge route.
The product can be narrower still. The Aave v3 annex says the cover “exclusively applies to deposits of crypto assets” in listed markets, which leaves a borrower’s position out of it. For events of the oracle or liquidation type, the loss counted is the one in any two-hour period.
Payout conditions: proof, timing and recoveries
When you file, you must be a member with KYC complete, outside the restricted list, and able to give cryptographic evidence linking you to the affected account. The claim must come after the cool-down and during the cover period or the 35-day grace period after it. Under the Crypto Cover terms, the grace period is 120 days for custody, and a depeg claim must come within 30 days of the trigger being met. Losses are computed at the time they occurred, minus any reimbursements, using CoinGecko rates or another reputable pricing site.
After a payout, the member has 20 days to assign rights to any reimbursement, and five days to repay anything received later. You cannot collect from cover and from a compensation plan for the same dollars.
Capacity and the provider’s own risk
Capacity is the limit a listing title leaves out. On 1 November 2026 Aave v3 had about 8.4 million dollars of room, Uniswap v3 about 8.8 million, and those numbers cannot be added up because the same staked NXM backs several listings. Nexus’s minimum capital requirement is active cover divided by a gearing factor of 4.8, about 15.7 million dollars on the 1 November numbers, so the pool is roughly six times that floor.
A comfortable margin in a normal year says little about a correlated one. If one bug hit a dependency shared by several covered protocols, claims would arrive together and fall on stakers and the pool at once. The pool can lose money on its own investments, and the mutual’s contracts are code that can fail as well. Nothing here says that will happen; it is the reason to count the provider among your risks.
Depeg and custody covers: severe triggers
Depeg cover is the best example of a title that promises more than the text. The Sky USDS annex on Sky‘s dollar pays only if the token trades more than 10 percent below its reference for seven consecutive days, and you must have held it for at least 72 hours before the event. You transfer the depegged token to Nexus before the claim goes to assessment; if approved, you receive 0.975 per token, and if denied, you get the token back. A dip to 95 cents for three days pays nothing, and neither does a two-day fall to 70 cents that recovers. A seven-day trigger is written for a lasting stablecoin collapse, not for a scare.
Custody cover has the same shape. Under the Crypto Cover terms it responds if the custodian is the victim of a crime and announces that all users must bear a loss of 10 percent or more, or if it imposes a complete withdrawal freeze without notice that lasts 100 days or more. On the freeze route the cool-down is itself 100 days, so a freeze of 99 days is outside it.
Kelp and Aave in 2026: who pays when the hole is big
On 18 April 2026 a forged message on Kelp’s bridge route released about 116,500 rsETH, a loss of 293 million dollars on DefiLlama’s count, as the hacks lesson recounts. Most of it went straight to Aave as collateral. The cover question is what stood between that hole and depositors, and the Aave incident report of 20 April lays it out:
| Item | Figure, 20 April 2026 |
|---|---|
| rsETH deposited on Aave | 89,567 (about $221M) |
| Bad debt, losses socialised across all rsETH | About $123.7M (scenario) |
| Bad debt, losses on L2 rsETH only | About $230.1M (scenario) |
| Umbrella WETH module | About $54M, of which most had begun a 20-day cooldown |
| Aave DAO treasury | About $181M |
The backstop built for this job was two to four times smaller than those scenarios, and it protects one asset on one network. Umbrella “wasn’t used”, Cointelegraph reported on 7 May. DeFi United, a coalition led by Aave, raised over 300 million dollars in ETH, according to The Block on 12 May. A governance proposal of 9 May put recovered rsETH at 106,993, and said the roughly 5,211 rsETH the attacker had sold on exchanges was to be covered by the coalition’s committed ETH.
Those are the dated facts as of mid-May 2026. A collection among protocols filled a gap that the staked backstop did not touch, and no rule obliges anyone to repeat it.
Cost against probability: what the price is really saying
A worked example with the 1 November prices
This is an example with fictional holdings and real quotes of 1 November 2026. It is arithmetic, and it says nothing on whether anyone should buy.
- A reader has 10,000 USDC supplied on Aave v3 and pays 110 dollars for a year of cover on that amount.
- After the 5 percent deductible, or 500 dollars, a total covered loss pays at most 9,500 dollars, and a 400 dollar loss pays nothing.
- Break-even probability: 110 divided by 9,500 is about 1.16 percent. Below that yearly chance of a covered total loss, the premium costs more than its expected payout.
- The Aave USDC supply rate on Ethereum was about 3.91 percent that morning, per DefiLlama, or 391 dollars a year. Cover takes about 28 percent of it, leaving about 2.8 percent. The one-off membership fee, about 4.50 dollars at that ETH price, comes on top.
The same reader holds 10,000 USDS with depeg cover at 50 dollars. The headline break-even is 50 divided by 9,750, about 0.51 percent. That flatters the product, because in a claim you hand over the token. If USDS sat at 80 cents after seven days, the gain from cover would be 17.5 cents per token, or 1,750 dollars on the position, and the break-even would be about 2.9 percent a year.
Together the two covers cost 160 dollars a year on 20,000, or 0.8 percent of the amount. A person may still pay more than the expected value for a loss they could not absorb, the way a household insures a house. That is a framing of the choice and no answer to it. Size matters too: a 10 million dollar order on Aave v3 would have exceeded the 8.4 million of room that morning.
What history says, and why it is not a probability
Here is DefiLlama‘s record for DeFi protocols, summed on 1 November 2026. The dataset gets revised.
| Year | Incidents | Total lost ($M) | Incidents of $10M or more |
|---|---|---|---|
| 2021 | 116 | 2,466 | 35 |
| 2022 | 141 | 2,852 | 24 |
| 2023 | 148 | 818 | 8 |
| 2024 | 160 | 796 | 13 |
| 2025 | 119 | 757 | 9 |
| 2026 to 1 Nov | 232 | 1,188 | 14 |
It is tempting to divide. In 2025 there were nine incidents of 10 million dollars or more, and on 1 November 2026 DefiLlama lists 203 protocols with 100 million dollars or more of TVL, leaving out centralised exchanges and chains, which gives about 4.4 percent. That number is wrong as a probability: the incidents hit protocols of every size, the two dates differ, the same protocol can be hit twice, and an incident rarely means a total loss for each depositor.
Nobody publishes a clean yearly chance of loss for a given protocol. A quote of 1.10 percent on Aave v3 and 4 percent on Curve All Pools is the closest thing to a market view, and it is set by stakers’ appetite and spare capacity as much as by risk. Curve’s listing covers many pools, so the two prices do not compare like for like.
Legal status: no deposit guarantee stands behind DeFi cover
Two different questions get mixed up when people ask whether DeFi insurance is safe in law. The first concerns public safety nets, and the answer is no across the board for a DeFi loss.
- United States: the FDIC fact sheet says deposit insurance does not apply to crypto assets and does not protect against the failure of non-bank entities, or against theft or fraud.
- United States: an SEC staff FAQ of May 2025, reviewed in February 2026, says SIPC protection extends to claims for securities, and answers no for non-security crypto assets held at a member broker-dealer.
- United Kingdom: the FCA warns that you generally won’t have access to the Financial Ombudsman Service and won’t be protected by the FSCS if a crypto firm goes out of business. A 2023 FSCS page says the same of platforms that exchange or hold crypto.
- European Union: the joint warning of the EBA, ESMA and EIOPA of 6 October 2025 says MiCA protections are less extensive than for traditional products, and that “you will not benefit from compensation schemes”. Italy’s insurance supervisor, IVASS, published it the same day.
- France: the FGDR states that Bitcoin and other cryptocurrencies are outside the deposit guarantee, which is capped at 100,000 euros per customer per institution.
- Italy: the FITD guarantees bank deposits at member banks, up to 100,000 euros per depositor. A position in a DeFi contract is no bank deposit.
The second question concerns the cover itself. Its status rests on the words already quoted: not a contract of insurance, and a discretion to pay. Nexus’s own forum adds a detail from the professional market. A proposal of 21 August 2025 states that “regulators won’t give any credit to traditional insurance companies” for cover bought on Nexus, and offers optional licensing of staking pools as a way to reach them. Where DeFi fits under MiCA and US law is a topic for a later lesson of this course.
Questions to ask before paying for cover
These questions come from the sections above. They make no case for buying or skipping.
- Which exact wording and annex apply, and does it cover a depositor, a borrower or an LP?
- Is the loss I worry about in the covered events at all, or in clause 10?
- Would the bridge, front end, wrapped token or underlying protocol in my path fall outside the annex?
- What is the deductible, and what does a total loss pay after it?
- Is there capacity for my amount and my term today?
- Can I pass KYC from my country, with the address that holds the position or one I can link to it?
- Could I wait out 14 days of cool-down plus the vote without that money, and risk a 0.05 ETH deposit on a claim that might be denied?
- For depeg or custody cover, which trigger must be met, and for how long?
- If the provider’s own pool is hit in the same event, what stands behind my claim?
The next lesson of the course covers the risks no cover pays for: a personal security routine for phishing, approvals and keys.
