Skip to content
PROTOCOL RESEARCH

Curve risk

Deep, battle-tested stableswap AMM at $1.277B TVL, but crvUSD's backing has drifted away from user collateral: about 17% of circulating supply is over-collateralized LLAMMA debt and another 15% is PegKeeper-minted, leaving roughly two thirds traceable mostly to DAO-authorized minting, decided by a veCRV electorate in which Convex alone holds a reported majority and which lost its independent risk manager in July 2026.
CautionResearched Aug 5, 2026

Scorecard

strong: Audits & contractsAll four named firms re-verified this run against primary sources, not carried from July: Trail of Bits (CurveDAO, Jul 2020, and Curve.fi, Jan 2020, both listed in the trailofbits/publications index), Quantstamp (gauges Aug 2020, metapool Oct 2020, certificate.quantstamp.com live), ChainSecurity (Curve Stablecoin assessment PDF live) and MixBytes (crvUSD report, Jun 2023, in mixbytes/audits_public). Deployed pools immutable once live. Curve's own docs audit index returned 404 for the second month running.
weak: Admin controlThe DAO can mint crvUSD with no collateral behind it and has done so repeatedly (a $60M credit line to Yield Basis, run by Curve's founder, plus 5M into Resupply's sreUSD market). DeFiScan rates Curve Stage 0 on this basis. Proposal #1252 (passed 12 Nov 2025) also gave an Emergency DAO multisig shared admin over crvUSD and LlamaLend risk parameters, a fast path around the 7-day vote plus 3-day timelock. New this window: risk manager LlamaRisk departed and the DAO opened a call for replacements on 8 July 2026.
strong: OracleInternal pool EMA oracles for the AMMs; crvUSD and LLAMMA price collateral off an aggregated on-chain feed with a Chainlink sanity bound. No oracle incident this window, and the peg read $0.99926 today, about 7bp off par. The older ChainSecurity flash-loan finding on the aggregate-price contract was medium severity and was addressed.
strong: Liquidity & exit$1.277B TVL, up about 4.7% since the 2026-07-01 report, with $1.211B of it on Ethereum. Exit from the main pools is normally a same-block swap. One named exception: the CRV-long LlamaLend market sits at 100.0000% utilization ($2.96M debt against $2.96M assets), so lenders there have no idle liquidity to withdraw against.
watch: Yield (real vs emission)Base LP yield is real swap fees, and crvUSD borrowers pay interest that funds the scrvUSD savings rate. But a large share of headline gauge APR is CRV emissions directed by veCRV votes, which inflate the number and decay. Underwrite the fee yield and treat CRV as a variable top-up that can be voted away.
watch: Holder concentrationConcentrated, but not in the direction the July report implied. Secondary reporting puts Convex at about 52.7% of veCRV voting power, Stake DAO about 14.7% and Yearn about 10.5%. The founder does not command this electorate: his own 17.45M CRV (about $6.2M) grant to Swiss Stake was rejected in December 2025, 54.46% against, with Yearn and Convex addresses making up roughly 90% of the opposition. The live risk is that a few liquid-locker protocols optimizing for veCRV cash flow now set crvUSD risk parameters. Precise current split: _refresh_.
watch: Track recordJuly 2023 Vyper compiler exploit drained Curve pools (~$73.5M across protocols, ~73% returned); the ~$10M June-2024 founder-loan bad debt is confirmed fully repaid, but a separate ~$700K CRV-long hole from the October 2025 crash is still open and that market still reads 100% utilized today. Recurring 2025 front-end attacks (X account, DNS) are a distinct and live surface. No exploit, depeg or fund loss found between 2026-07-01 and 2026-08-05.
🟢 strong🟡 watch / caveat🔴 weak / fund-loss risk
Verdict is a gate (worst flaw wins), not an average. Our read, not financial advice.

Live facts

auto-sourced now
TVL$1.30B
30d↑2%
Audits2
Last hack$240k · Mar 2026

DeFiLlama + our exploits feed. Cross-check the dated report against today.

TL;DR

Curve is still the reference stableswap AMM and it grew this month, to roughly $1.277B TVL. The refresh turned up one thing that changes the shape of the risk rather than its size. crvUSD is described nearly everywhere, including in our own dependency map, as an over-collateralized stablecoin with soft-liquidation bands. Today that describes a minority of it. Of $228.27M crvUSD circulating, about $38.2M is debt against the nine LLAMMA mint markets, collateralized at a healthy 191%, and a further $33.75M is PegKeeper debt, crvUSD the protocol minted straight into the crvUSD/USDT pool. Together those two documented mechanisms account for roughly 31% of circulating supply. The remaining two thirds traces predominantly to DAO-authorized minting, including up to $60M approved as a credit line for Yield Basis, a protocol run by Curve's own founder, and 5M minted directly into Resupply's lending market under a model the DAO itself described as using lending pairs "rather than traditional collateralized minting pairs." That is a governance power working as designed, not a bug. It is still the reason this stays at caution: the backing of a stablecoin now depends materially on who the DAO lends newly minted supply to. On the old number, there is real good news. The roughly $10M of founder-loan bad debt from June 2024 is confirmed fully repaid. A smaller and newer hole, about $700K in the CRV-long LlamaLend market from the October 2025 crash, is still open.

Checklist

Audits & contracts. Curve carries multiple Tier-1 audits across core and crvUSD, and this run they were re-verified against the firms' own primary records rather than carried forward. Trail of Bits lists both a CurveDAO review (July 2020, six findings) and an earlier Curve.fi review (January 2020) in its own publications index. Quantstamp's certificate for the gauge and metapool work resolves live. ChainSecurity's Curve Stablecoin assessment PDF resolves live. MixBytes' crvUSD report sits in its public audits repository. No invented firms. Deployed pools are largely immutable, which removes a whole class of upgrade risk. The 2023 loss came from a Vyper language bug underneath the audited source, a standing reminder that an audit assumes the compiler is sound. One honest gap: Curve's own docs audit index at docs.curve.finance/references/audits returned 404 for the second month running, as did every asset path under it, so the firm list is verified from the auditors, not from Curve.

Admin control. This is the dimension that moved, and it moved to red. The Curve DAO can mint crvUSD with nothing behind it, and reassign control of key contracts. DeFiScan rates Curve Stage 0 on its decentralization framework for exactly that reason. This is not theoretical. The DAO approved a credit line of up to $60M crvUSD, implemented as a pre-mint, to Yield Basis, a protocol built by Curve's founder, over objections that nobody independent had priced the economic risk and that a hack there would leave Curve holding the liability. A separate 5M crvUSD mint went into Resupply's sreUSD LlamaLend market in October 2025, explicitly introducing a minting model that bypasses collateralized mint pairs. On top of that, proposal #1252 passed on 12 November 2025 and gave an Emergency DAO multisig shared admin over crvUSD and LlamaLend risk parameters: debt ceilings, AMM fees, monetary policy, loan discounts, vault deposit caps. Ordinary changes still run 7 days of voting plus a 3-day timelock. Emergency ones do not. Those powers are bounded to risk parameters and are mostly risk-reducing rather than fund-seizing, which is the fair counterweight. What is new and unhelpful is that the DAO's independent risk assessor, LlamaRisk, left, and on 8 July 2026 the DAO opened a two-week call for replacement teams to monitor crvUSD and LlamaLend. Discretionary minting powers are easier to live with when someone independent is pricing them.

Oracle. Unchanged and still fine. AMM pools use internal EMA price oracles, and crvUSD's LLAMMA engine prices collateral off an aggregated on-chain oracle with a Chainlink sanity bound. No oracle incident surfaced in this window, and the peg read $0.99926 today, roughly 7bp off par, which is normal noise for this asset. The older ChainSecurity flash-loan finding on the aggregate-price contract was medium severity and was addressed.

Liquidity & exit. At about $1.277B TVL, up roughly 4.7% from the $1.22B in the July report and with $1.211B of it on Ethereum, Curve remains one of the deepest venues in DeFi for stablecoin and liquid-staking pairs, and exit is normally a same-block swap. One exception deserves naming rather than averaging away: the CRV-long LlamaLend market currently shows $2.96M of debt against $2.96M of assets, which is 100% utilization to four decimal places, so lenders in that specific market have no idle liquidity to withdraw against. That is the same market carrying the open bad debt. The general caveat also stands. In a coordinated depeg the same pools that give you a cheap exit are the ones that concentrate the loss.

Yield: real vs emission. The base layer is real, being LP fees from swap volume, and crvUSD borrowers pay interest that funds the scrvUSD savings rate. But a large part of the advertised APR on many gauges is CRV emissions directed by veCRV votes, and emissions inflate the headline and decay. Size the position on the fee yield. Treat CRV rewards as a top-up that can be voted away.

Holder concentration. Still the soft spot, but the July framing of it was wrong and is corrected here. veCRV power is not concentrated in the founder. Secondary reporting puts Convex at roughly 52.7% of veCRV voting power, Stake DAO near 14.7% and Yearn near 10.5%, so the liquid-locker protocols, not Swiss Stake, decide outcomes. The proof ran in December 2025, when the founder's own request for 17.45M CRV (about $6.2M) to Swiss Stake AG was voted down, 54.46% against to 45.54% in favour, with Yearn and Convex addresses making up close to 90% of the opposition. So the $60M Yield Basis line did not clear because the founder controls the vote. It cleared because the large lockers were persuaded it would route fees to veCRV. That is the real exposure to underwrite: a handful of vote-aggregating protocols, whose mandate is veCRV cash flow rather than crvUSD holder safety, now set the minting and risk parameters, and as of July they do so without an independent risk assessor at the table. Precise current voting-power split is unverified this run.

Track record. Several distinct events, and they should not be blurred together. The July 2023 Vyper exploit was a genuine contract-level exploit of Curve pools, roughly $73.5M across affected protocols with about 73% returned within a week. That is separate from the 2025 front-end attacks, where the X account and the curve.fi DNS were hijacked in May for phishing and no contract funds were lost; Curve has since frozen the old domain and moved to a new one. Separate again is the founder-loan bad debt, and here the record improved: the roughly $10M left on LlamaLend by the June 2024 liquidation cascade is confirmed fully repaid, not socialized. What replaced it is smaller and messier. The October 2025 crash left the CRV-long LlamaLend market with a collateral shortfall near 70% and about $700K of bad debt, and as of the April 2026 proposal the plan was not repayment but a market mechanism: package the underwater positions into vault tokens and sell claims at a discount through a dedicated StableSwap pool priced near 71% of face. That shifts the loss to whoever buys the claim instead of erasing it, and today's read of that market at 100% utilization is consistent with it still being impaired. No exploit, depeg or fund loss surfaced between 2026-07-01 and 2026-08-05. The two developments in the window were structural rather than破: the LlamaRisk departure and the LlamaLend v2 rollout that began on Optimism on 10 June 2026 with mainnet planned later in the year.

Worst case

A partner protocol holding a large pre-minted crvUSD credit line blows up, and the crvUSD it borrowed is already circulating with nothing recoverable behind it. Yield Basis is the obvious candidate, since it holds the largest approved line, is run by the same founder, and had community objections on record that Curve would end up wearing the liability. Because that supply is not a LLAMMA position, no soft-liquidation band unwinds it. The DAO absorbs it, or crvUSD holders do through the peg. A quieter version: the large lockers keep approving mints that route fees to veCRV, backing quality drifts down gradually, and with no independent risk team seated since July nobody flags it until a stress window. On the periphery, another DNS or front-end hijack tricks you into signing a malicious approval, which is an operational loss rather than a protocol failure. Curve has no protocol-wide insurance fund covering LPs, so position sizing and bookmarked, verified contract addresses are the mitigations.

Bottom line

Caution, same verdict as July, sharper reason. The AMM core is deep, heavily audited and mostly immutable, the peg is holding, TVL is growing, and the old $10M bad debt is genuinely cleared. What keeps it off solid is no longer mainly the 2023 scar tissue. It is that crvUSD's backing has quietly shifted from user CDPs toward discretionary DAO minting, decided by a veCRV electorate that a few liquid lockers dominate, with a related-party allocation already approved, an emergency multisig able to move risk parameters fast, and no independent risk manager in the seat since July. LPing deep Curve stable pools is a different and much smaller risk than holding size in crvUSD. Treat those as two decisions, not one, and re-read the backing mix before sizing the second.

Dependency-map note

Our map carries one edge for Curve labelled "crvUSD backed by over-collateralized crypto positions with soft-liquidation bands." As of today that label is only partly accurate and should be amended. Verified this run: $228.27M crvUSD circulating, of which about $38.2M (roughly 17%) is debt across the nine LLAMMA mint markets, collateralized at about 191% ($73.1M of collateral), and a further $33.75M (roughly 15%) is PegKeeper debt held by a single active keeper on the crvUSD/USDT pool, which is protocol-minted stabilization inventory rather than a user CDP. The remaining roughly 68% traces predominantly to DAO-authorized minting: pre-mint credit lines to partner protocols, and crvUSD minted directly into LlamaLend vaults. Note that a portion of that residual is collateralized once borrowed, since LlamaLend borrowers post collateral: crvUSD lending vaults on Ethereum hold $54.93M with $43.74M lent out against collateral and $11.18M idle. So the correct amendment is "partly over-collateralized user positions, partly protocol-minted, and substantially DAO-minted," not simply "unbacked."

Data appendix

  • TVL: ~$1.277B (DeFiLlama, slug curve-dex, $1,277,085,785), up ~4.7% from the ~$1.22B recorded 2026-07-01; Ethereum $1,210,504,676. Curve LlamaLend is a separate slug and is not included here. 90d and ATH trend: refresh.
  • crvUSD: price $0.99926, circulating $228,274,030 (DeFiLlama stablecoins API, id 110); Ethereum share $226,144,189. On-chain totalSupply of the crvUSD contract reads $2,089,809,205, which is minted-into-controllers headroom rather than circulating supply, and is the pre-mint architecture itself showing up in the numbers.
  • crvUSD backing composition: nine LLAMMA mint markets on Ethereum, total debt $38,221,972 against $73,073,417 collateral (191%), largest being WBTC $15.38M, wstETH $10.50M, cbBTC $4.51M, tBTC $4.47M, WETH $1.75M (Curve prices API, /v1/crvusd/markets/ethereum). PegKeepers: 15 deployed, 5 active, total debt $33,750,000, all of it in the single active crvUSD/USDT keeper 0xFb726F57d251aB5C731E5C64eD4F5F94351eF9F3 (/v1/crvusd/pegkeepers/ethereum); that pool currently holds $20,916,971 crvUSD on-chain. LLAMMA plus PegKeeper is 31.5% of circulating; the ~$156.3M residual (68.5%) is attributed directionally, not line by line: refresh.
  • LlamaLend: 51 Ethereum markets, total debt $43,743,205 against $54,942,224 assets (Curve prices API, /v1/lending/markets/ethereum). crvUSD-borrow vaults specifically: $54,927,225 supplied, $43,743,215 lent out, $11,184,010 idle. CRV-long market: $2,964,968.99 debt against $2,964,969.05 assets, i.e. 100.0000% utilization.
  • Audits: re-verified this run from primary sources, not carried from July. Trail of Bits: CurveDAO (Jul 2020, 6 findings) and Curve.fi (Jan 2020), both listed in trailofbits/publications. Quantstamp: Curve gauges/VestingEscrow (Jul-Aug 2020) and metapool (Oct 2020), certificate.quantstamp.com resolves. ChainSecurity: Curve Stablecoin code assessment, reports.chainsecurity.com PDF resolves. MixBytes: crvUSD Security Audit Report (5 Jun 2023) and Curve DAO Voting, in mixbytes/audits_public. Curve's own index docs.curve.finance/references/audits returned HTTP 404 again, as did all /assets/pdf/audits/ paths tried.
  • Admin / governance: Curve DAO via veCRV, 7-day on-chain voting plus 3-day parameter timelock; deployed pools immutable. Emergency DAO multisig granted shared admin over crvUSD and LlamaLend risk parameters via proposal #1252, announced 5 November 2025, voting closed 12 November 2025 (PANews, MEXC News, Phemex). DAO can mint crvUSD without collateral and reassign key contracts; DeFiScan rates Curve Stage 0 on this basis, per DeFiScan builder TokenBrice, who called the Yield Basis proposal a mint of "out-of-thin-air, unbacked, 60M crvUSD" (Blockworks). $60M crvUSD credit line to Yield Basis approved, implemented as a pre-mint, pools initially capped at $10M each across WBTC, cbBTC and tBTC (Cointelegraph, Blockworks, Curve gov forum thread 10774). Separate 5M crvUSD mint into Resupply's sreUSD LlamaLend market, proposal 1237, announced 24 October 2025, explicitly using "lending pairs rather than traditional collateralized minting pairs" (Curve news blog). Risk manager LlamaRisk departed; DAO call for replacement risk teams published 8 July 2026, two-week application window. Primary on-chain records for #1252 and the DeFiScan page itself: refresh.
  • Oracle: internal pool EMA oracles; crvUSD and LLAMMA use an aggregated on-chain price with a Chainlink sanity bound (Curve docs, ChainSecurity crvUSD audit, carried from July). Peg verified live at $0.99926.
  • Holder / governance concentration: secondary reporting puts Convex at ~52.7% of veCRV voting power, Stake DAO ~14.7%, Yearn 10.5%, with roughly 40-50% of circulating CRV vote-locked; not verified against a primary source this run. Founder relocked veCRV to 2029. December 2025: Egorov's 17.45M CRV ($6.2M) grant request for Swiss Stake AG was REJECTED, 54.46% against to 45.54% for, Yearn and Convex ~90% of the opposing vote (The Defiant, KuCoin, ainvest, PANews).
  • Related exposure: Yield Basis TVL $132,053,257 (DeFiLlama, slug yield-basis). Amount of the $60M line actually drawn today: refresh.
  • Recent-news scan (2023 to 2026-08-05): July 2023 Vyper exploit, ~$73.5M across pools, ~73% returned (Chainalysis, Rekt, Blockworks, carried from July). Aug 2023 brief crvUSD depeg ~0.35%, contracts unaffected. June 2024 founder liquidation left ~$10M bad debt on LlamaLend, confirmed fully repaid within days (Cointelegraph). Dec 2024 second founder liquidation. May 2025 X-account (May 5) and DNS (May 12) hijacks, front-end only, old domain frozen and a new one adopted (The Block, Cointelegraph, Decrypt). Oct 10 2025 crash left ~$700K bad debt and ~70% collateral backing in CRV-long; April 2026 market-based recovery proposal using tokenized vault claims in a StableSwap pool (~1% redemption fee, priced near 71% of face), vote outcome not confirmed (CoinDesk, crypto.news, BanklessTimes, gov forum thread 11062). Sept-Nov 2025 Yield Basis credit line and Emergency DAO votes. Dec 2025 Swiss Stake grant rejected. June 2026 LlamaLend v2 launched on Optimism with a 250,000 OP incentive, mainnet planned later in the year (The Block, crypto.news). July 2026 LlamaRisk departure and risk-team call for proposals. No Curve exploit, front-end incident or crvUSD depeg found between 2026-07-01 and 2026-08-05. Maintained monthly. Methodology: DeFi Research Instruction v2.

← all protocols