Kelp
Tier 2Security Analysis Report
TVL
$663.04M
Sections
7
report sections
Citations
610
evidence references
Version
v16
11/13/2026
Moderate Risk
Safety Score · out of 10
Executive Summary
Executive Summary: Kelp
Kelp's core staking design on Ethereum is the stronger part of the protocol. The weak points are the bridges that carry rsETH to other chains, thin market liquidity, and Safes (shared multi-signature wallets) that change key settings with no delay. A bridge failure cost about $293M in April 2026. Most evidence about the recovery comes from weak publishers, so it is reported, not confirmed.
Safety Rating
Overall Score: 4.9/10 Dimension scores: Code security 5.0 · Economic security 4.9 · Team transparency 5.7 · Governance 3.7 · Track record 4.9
- Code security
- 5.0 points
- Economic security
- 4.9 points
- Team transparency
- 5.7 points
- Governance
- 3.7 points
- Track record
- 4.9 points
The cross-analysis judges the core Ethereum mechanism (balance-based rate and a 10-day upgrade delay) sounder than the layer around it, set against a realised bridge loss, thin exit liquidity and instant admin powers.
Scores by Dimension
| Dimension | Key Finding |
|---|---|
| Code Security | Several independent firms reviewed the core contracts, but no deployed contract is shown to match an audited version, and most bridge, pool and Safe contracts have no confirmed audit scope. |
| Economic Security | The rsETH rate is computed on-chain from held balances, but the token is thinly traded, depends on bridges and other protocols, and no Kelp-owned loss fund is documented. |
| Team Transparency | Two co-founders and a company are named, but the company's jurisdiction and the founders' prior work are not independently verified, and public code activity is sparse. |
| Governance | Control sits in a few Safes with a 10-day timelock on upgrades and asset onboarding only; no binding vote was found, and several fund-affecting powers act instantly. |
| Track Record | Operating since late 2023, with three recorded incidents: the April 2026 bridge exploit and a 2024 website attack concern this deployment, and the third is a DefiLlama record for the family product Gain, whose link to Kelp Gain is not independently established; how long exits stayed halted afterwards rests only on unverified-publisher sources. |
What This Protocol Does
Kelp lets people deposit ETH or staked-ETH tokens and receive rsETH, a token that represents their share of the pooled deposits. Kelp restakes the deposits on EigenLayer, where staked ETH also secures other services for extra rewards.
The DefiLlama record holds one product, Kelp, which issues rsETH. TVL (total value locked, the assets deposited) is about $739M, all on Ethereum, down from a peak above $1.8B. Gain and Kernel are other products of the same team. They sit outside this record and are not assessed.
Key Strengths
- Rate design (rsETH on Ethereum). The rsETH rate is computed from held balances, not a market price, with a 1% per-update bound. The docs say a single swap or flash-loan sandwich in one block cannot move it.
- Review depth (core deposit, oracle and withdrawal contracts). MixBytes, Sigma Prime, Bailsec and a Code4rena contest (source reliability: unverified publisher) reviewed these contracts between 2023 and 2026. This credit does not extend to the bridge verification path.
- Public source and bounty (core Ethereum contracts). Source is public for the main Ethereum contracts, and rsETH is covered by an Immunefi bounty of up to $250,000.
- Upgrade delay (core Ethereum proxies). Core upgrades pass through a 10-day timelock, confirmed on-chain at block 26379467 (2026-11-09). The standard withdrawal takes 15-21 days, so a holder who sees a harmful upgrade queued cannot finish a standard withdrawal before it executes.
- Backing is checkable (rsETH on Ethereum). Backing is the on-chain balance of deposit assets, summed by a proof-of-reserves reader. The check rests on Kelp's own reader contract.
Key Risks
- 1. Bridged rsETH: a repeat forged message could empty the lockbox behind a route. Magnitude: the April loss was $293M, 18.41% of the previous day's TVL. The two Ethereum lockboxes hold about 8.5% of Ethereum supply, roughly $63M. Likelihood: medium. Reading the exploited route's current settings and the CCIP risk-management settings would move the likelihood to low or high.
- 2. rsETH on Ethereum: a thin market and one dominant holder can impair exit. About 84% of supply sits in one proxy, which the economic report infers (not confirmed) to be Aave's rsETH token. Magnitude: loss of access, not loss of principal, with about 286,400 rsETH (about $625M) in one market if the inference holds. Likelihood: high that market exit is impaired in stress, medium that a pause or freeze recurs. A 1-of-5 Pauser Safe can halt exits with no stated time limit.
- 3. Instant Safe powers bypass the 10-day timelock. A compromised or colluding Safe could change fund-critical settings at once. Magnitude: the CCIP lockbox is 3.55% of supply (about $26M); the worst case is theoretical until the on-chain bound on registry changes is read. Likelihood: low.
- 4. The bridge verification layer has no confirmed audit coverage. An April incident account says the verification layer sits outside every audit scope, and the collected audits address contract code, not the off-chain verifier that failed in April. Coverage of that layer is unconfirmed, not shown absent. The May 2026 Bailsec report covers a configuration script only.
- 5. No Kelp-owned loss fund, and recovery is unconfirmed. No Kelp insurance, reserve or compensation fund appears in the sources. The refill accounts all come from weak sources, and the adapter balance was not checked on-chain.
Critical Findings
Audit findings
Counts are issue groups, protocol-wide, as published in the code report. The last column is audit status, not a count of confirmed live vulnerabilities. Audit coverage is partial: several reports sit on hosts not verified as the auditor's own, and the Bailsec May 2026 review covers an off-chain script, not smart contracts.
| Severity | Total groups | Fixed per the auditor | Not marked fixed by the auditor |
|---|---|---|---|
| Critical | 3 | 3 | 0 |
| High | 22 | 14 | 8 |
| Medium | 45 | 18 | 27 |
No critical group is marked other than fixed. Audited code was not compared with deployed code, so equivalence between the two is not established. The Check results below are a separate, direct comparison of each finding with verified deployed source where available. A visible fix in the repository does not prove deployment, and a Check never changes an audit status. A result of described code still there is a limited signal, not a confirmed vulnerability.
High groups not marked fixed (8):
- Possible arbitrage from Chainlink price discrepancy (Code4rena contest, source reliability: unverified publisher). Audit status: disputed. Check: the checked version of the code does not contain the described code in this form (deployed-code check).
- First-staker rsETH price manipulation (Code4rena contest, source reliability: unverified publisher). Audit status: acknowledged. Check: the checked version of the code does not contain the described code in this form (deployed-code check).
- The reward calculation may be blocked until contract upgrade (MixBytes, March 2024). Audit status: acknowledged. Check: unavailable, because the audited repository code was not available.
- Stale rsETH price and yield extraction (MixBytes, Sigma Prime and Bailsec.io). Audit status: acknowledged. Check: the described code is unchanged in the verified deposit pool source (deployed-code check, Sourcify runtime match).
- Price changes result in stuck tokens (Sigma Prime, November 2024). Audit status: acknowledged. Check: unavailable, because the audited code was not found in the collected files.
- Daily fee limit blocks price updates (Bailsec.io, September 2025). Audit status: acknowledged. Check: the described code is unchanged in the verified oracle source (deployed-code check, Sourcify runtime match).
- Hub-and-spoke topology is assumed by iteration order, not verified (Bailsec, May 2026). This concerns the off-chain verification script, not a deployed contract. Audit status: partially fixed. Check: unavailable, because the audited code was not found in the collected files.
- First depositor can brick the protocol via rounding (SBSecurity analysis, source reliability: unverified publisher; not an audit). Audit status: open. Check: the checked version of the code does not contain the described code in this form (deployed-code check).
Medium groups not marked fixed (27), by our Check result:
- Described code unchanged in verified deployed source (11 groups). Audit status for each: acknowledged, except one not stated and one open; one is an estimated severity.
- Skewed price calculation if future LSTs use decimals other than 1e18 (Code4rena; audit status not stated).
- Protocol functionality lockout upon adding a new token (MixBytes).
- NodeDelegator index shuffling and frontrunning (Bailsec.io, Sigma Prime).
- FeeReceiver balance not included in TVL (Sigma Prime).
- stETH 1-2 wei corner case (Bailsec.io).
- Missing staleness verification in
getAssetPrice(Bailsec.io). sweepRemainingAssetsmay be blocked indefinitely (Bailsec.io).- User can hedge slashing losses with an instant withdrawal (Bailsec.io).
maxAmountToDepositBridgerAssetcounts only one alternative rsETH token (Bailsec.io); the check used Etherscan-listed source, an explorer claim.- LRTManager can pause minting by setting the daily mint cap to 0 (Bailsec, December 2025): Medium (estimated), our assessment because the audit states no rating.
- High centralization risk from LRTManager and LRTAdmin roles (SBSecurity analysis, source reliability: unverified publisher; audit status open).
- Described code not found in this form (7 groups, deployed-code check). Arbitrage in the rsETH price calculation (MixBytes); centralization risks (MixBytes);
getTotalAssetDepositsincorrect value for ETH (Bailsec.io); storage collision in LRTOracle upgrade (Bailsec.io); instant withdrawals can brick the withdrawal queue (Bailsec.io); rsETH not compatible with CCIP self-service registration (Bailsec.io); stETH rebase can hit deposit limits (SBSecurity analysis, audit status open). All are acknowledged except the last. - Fix visible in verified deployed source (4 groups). Strategy update without migration of deposits (Code4rena, Rezolv; disputed); lack of slippage control on
depositAsset(Code4rena, Rezolv; audit status not stated); Ether cannot be withdrawn from NodeDelegator under certain conditions (MixBytes; acknowledged); no way to remove deprecated supported assets (SBSecurity analysis; open). - Check unavailable (2 groups). Unsafe contract deployment and update process; the possible inability to initialize the eigenPod field (both MixBytes, acknowledged); both lack audited repository code.
- Repository-only check (2 groups). Incorrect
stakedButUnverifiedNativeETHaccounting (Sigma Prime; acknowledged) and no withdrawal functionality from EigenLayer (SBSecurity analysis; open). The default-branch snapshot does not contain the described code in this form, and deployment is not confirmed. - Inconclusive (1 group). ChainlinkPriceOracle uses a stale
latestAnswerwithout safety checks (SBSecurity analysis; open). The finding nameslatestAnswerwhile the verified source callslatestRoundData, so no verdict is supported.
Fixed audit groups appear only in the counts above.
Other material issues from the reports
- Realised bridge exploit (2026-04-18). Kelp's own post says a forged packet released 116,500 rsETH from the Ethereum bridge lockbox. DefiLlama records $293M. Responsibility is disputed between Kelp and LayerZero and neither account is independently confirmed.
- Instant admin paths. A 3-of-6 Safe controls mint limits, withdrawal fees, user transfer blocking and bridge verifier configuration without delay. The signers of the Safes are not identified, and the claim of external signers is unconfirmed.
- Thin market and concentration. About USD 0.5 million of depth at 2% price impact against USD 737 million of supply. The top holder owns 84.09% of Ethereum rsETH.
Documented versus on-chain
- Kelp's docs (undated) say every LayerZero route enforces 4-of-4 verifiers. On-chain on Ethereum at block 26393314 (2026-11-12) the route to Arbitrum reads 4 of 4 with 64 confirmations. The exploited Unichain-to-Ethereum route has no read, so "all routes" is not measured.
- The same route can be changed with no delay.
- Kelp's docs (undated) say timelock proposers are the "respective admin multisigs". On-chain on Ethereum at block 26379467 (2026-11-09) the proposer is the 6-of-11 Safe, not the 3-of-6 Admin Safe.
- Kelp's docs page lists LayerZero for Base, Ink and Mantle, while a Kelp post dated 2026-10-12 says those moved to Chainlink CCIP on 2026-10-05. The two have not been reconciled.
- Kelp's docs state a 0.5% instant-withdrawal fee. On-chain at block 26379467 (2026-11-09) the raw value is 50, which equals 0.5% only if the deployed contract uses a 10,000 denominator, which is not verified.
Recommendation
Assessment: Kelp suits risk-tolerant users who accept bridge, exit-liquidity and admin-key risk, and who prefer the Ethereum mainnet route over bridged copies. The core Ethereum mechanism is judged sounder than the layer around it, but deployed code is not shown identical to an audited revision, and a realised $293M bridge loss, thin market depth, a pause power with no stated limit and instant Safe powers weigh against it. Users with low risk tolerance should wait for a read of the exploited route's current settings, on-chain confirmation of lockbox backing, and timelocks on the bridge and registry controls. Holders of old bridged copies on the networks in the recovery guide can recover them by paying a fee, until 2027-06-15. Holders of CCIP rsETH on Zircuit have no standard exit, because that token is outside the recovery program.
Report Sections
- Protocol Overview
- Code Security
- Economic Security
- Social & Team
- Governance
- Cross-Risk Analysis
83 citations · click markers to view evidence
Protocol Overview
Protocol Overview: Kelp
What It Does
Kelp lets people deposit ETH or staked-ETH tokens and receive rsETH, a token that represents their share of the pooled deposits. Kelp restakes the deposits on EigenLayer, where staked ETH also secures other services for extra rewards. rsETH rises in value against ETH as rewards accrue.
The Ethereum-side design is documented and its controls were read on-chain. Kelp reports that a forged bridge message released 116,500 rsETH in April 2026, so the main risk sits in the bridges rather than the staking logic. DefiLlama shows TVL (total value locked, the assets deposited) of about $739M on Ethereum, down from a peak above $1.8B.
- Category: liquid restaking. DefiLlama lists it as .
- Products in the DefiLlama record: one product, Kelp, issuing rsETH (a liquid restaked token, or LRT). Kelp's docs describe it as . DefiLlama counts TVL on Ethereum only .
- Other products of the same team (context only): Kelp's September 2025 post says KernelDAO unites Kelp, Gain (vaults) and Kernel (BNB Chain restaking) . Gain and Kernel are other products outside this DefiLlama record. This report does not assess them.
- Target users: ETH and liquid-staking-token holders who want restaking rewards while keeping a token they can lend, borrow or trade .
- Scope of the bridge layer: rsETH also exists as bridged copies on other chains. The bridge contracts, not the restaking contracts, were the point of failure on April 18, 2026 (see How It Works and Key Risks).
How It Works
Kelp is a pooled deposit system. Users deposit supported assets, a contract mints rsETH at an oracle rate, and operators restake the pooled assets. The user flow below is for Ethereum mainnet. Contract roles were read on-chain at block 26379467 (2026-11-09).
- The user connects a wallet to the Kelp app and chooses to restake ETH, ETHx or stETH . The on-chain list of supported assets has three entries, which matches .
- The user deposits into
LRTDepositPool. The deposit pool andLRTOraclehold the minter role onLRTConfig, the registry that stores roles and settings . ThersETHcontract caps minting at 5,000 rsETH per day . - Role-gated functions move the assets from the deposit pool to
NodeDelegatorcontracts (transferAssetToNodeDelegator). SevenNodeDelegatorcontracts were in the queue on-chain, with a limit of 20 . Node operators run validators and restake to EigenLayer services: The listed operators are P2P, Figment, Luganodes and Pier Two. LRTOraclesets the rsETH/ETH rate. It reads asset balances held across the protocol, prices them, sums them and divides by rsETH supply . Rewards raise the rate instead of paying out tokens.- On L2s, the deposit contracts read a rate that
RSETHMultiChainRateProvideron Ethereum relays through LayerZero V1 messages . Bridged rsETH moves through a lockbox on Ethereum (see the bridge description below). - To withdraw, the user picks an asset and a mode, and
LRTWithdrawalManagerhandles the request. Withdrawals happen on Ethereum only . Instant mode takes under 2 minutes and costs 0.5%, subject to available funds, while standard mode is free and takes 15-21 days . On-chain,instantWithdrawalFeereads 50 .
- (actor)
- (contract)
- (token)
- (oracle)
- (contract)
- (external)
- (contract)
- step 2flow
- step 2flow
- step 3flow
- step 3flowdependency
- step 4flow
- step 6flow
Core mechanism
rsETH is a reward-bearing token. The docs state . On-chain at block 26379467, rsETHPrice reads 1084187395548949112 . That is about 1.084 ETH per rsETH if the value uses 18 decimals, as the rsETH token does. The same read shows total supply of 345124439223188827996140 raw units, about 345,124 rsETH .
Rate guardrails
- A single oracle update may not move the price by more than
pricePercentageLimit, documented as 1% . The on-chain raw value is 10000000000000000 , which equals 1% on a 1e18 scale. - On a downside breach the oracle auto-pauses deposits, withdrawals and oracle updates .
- The rate is share accounting over held assets, not a market price. A proof-of-reserves reader recomputes it independently .
Bridge layer (cross-chain rsETH)
Kelp's own docs say any L2-to-L2 transfer routes through Ethereum . Value is locked in a lockbox on Ethereum, and a matching claim is minted or released on the other chain.
- Who verifies a message: on LayerZero routes, independent verifiers called DVNs (decentralized verifier networks). Kelp's docs now say . Community accounts say the route exploited on April 18, 2026 was configured as 1-of-1 (see below). Kelp's own incident post does not state this.
- Who relays: LayerZero and Chainlink CCIP infrastructure. Kelp's docs list and .
- Which contract releases value on Ethereum:
RSETH_OFTAdapter(LayerZero lockbox) and a CCIPLockReleaseTokenPoollockbox. - Two sources differ on the current routes. A Kelp post dated October 12, 2026 says It also says The Kelp docs page above still lists LayerZero for Base, Ink and Mantle. The same post says the new CCIP rsETH bridges as follows . Six networks now carry both tokens Assessment: the post separates old and new tokens, and the docs page does not, so the post is the more specific account. The two have not been reconciled.
- Network consolidation: Kelp says old rsETH bridging on 22 listed networks, including Optimism, Scroll, zkSync and Avalanche, was retired on June 15, 2026 . Holders of old rsETH can recover it by burning it and paying a fee until June 15, 2027 .
What happened on April 18, 2026
Kelp's own incident post says . It adds . DefiLlama records the event as $293M, technique "Cross-Chain Message Spoofing" .
- Mechanism (Aave service providers, community publication, April 20, 2026): the exploited route was . A forged packet released rsETH from the adapter. The adapter's balance fell from 116,723 rsETH to 223 rsETH. Of the amount released,
- Cause as LayerZero describes it (source reliability: community publication): The verifier then LayerZero's May 18, 2026 report (source reliability: unverified publisher) says
- Dispute over responsibility: LayerZero says Kelp chose a 1-of-1 setup despite its advice . CoinDesk (source reliability: unverified publisher) reports Kelp's counter-claim and notes Neither side's account is independently confirmed.
- Response: Kelp's emergency multisig froze core contracts 46 minutes after the drain, per Metrika (source reliability: community publication) . Chainalysis (source reliability: community publication) reports the Arbitrum Security Council froze 30,766 ETH .
- Open issue as of the April 20 Aave report: Later accounts describe the settlement, but all come from weak sources, so it is reported rather than confirmed.
- An Arbitrum forum proposal (source reliability: community publication) states
- An Aave governance post (source reliability: community publication) reports liquidations recovered rsETH . The same post says coalition funds were to cover the rest
- A news report (source reliability: unverified publisher) quotes a Kelp post on the last refill of the lockbox
- Litigation: an X post attributed to Kelp on September 25, 2026 (source reliability: unverified publisher) says .
Key smart contracts and roles (Ethereum)
| Contract | Role |
|---|---|
rsETH (proxy) | Receipt token; daily mint cap; per-address transfer blocking and fund recovery added in January 2026 |
LRTConfig | Role and settings registry; DEFAULT_ADMIN_ROLE holder is the External Admin Safe |
LRTDepositPool | Takes deposits, mints rsETH, forwards assets to NodeDelegator contracts |
LRTOracle | Computes the rsETH/ETH rate; bounded updates |
LRTWithdrawalManager, LRTUnstakingVault, LRTConverter | Withdrawals, unstaking and asset conversion |
NodeDelegator contracts | Hold restaked assets at EigenLayer |
RSETH_OFTAdapter | LayerZero lockbox on Ethereum |
CCIP LockReleaseTokenPool | Chainlink CCIP lockbox on Ethereum |
RSETHMultiChainRateProvider | Sends the rate to L2s |
The contract list is source-provided . The roles above come from on-chain reads .
Yield Sources
Yield comes from Ethereum staking rewards and EigenLayer restaking rewards, and it shows up as a rising rsETH/ETH rate. Kelp's docs say users earn . Kelp also earns Aave interest on idle withdrawal ETH . The sources do not establish how that interest affects the rsETH rate.
- Fees:
LRTConfigshowsprotocolFeeInBPSof 500 . That is 5% only if the deployed denominator is 10,000, which the available sources do not confirm.LRTOraclealso carries a fee-mint cap,maxFeeMintAmountPerDay, of 20000000000000000000 raw, or 20 rsETH on an 18-decimal scale . Instant withdrawals cost 0.5% . - DefiLlama fee data: . These are DefiLlama's dollar figures and its definition of fees is not stated in the files.
- Incentive points: rsETH depositors earn Kernel Points . Referral links add 10% extra points . Eigen programmatic reward claims were deprecated on September 23, 2026 .
- Leveraged demand: Kelp's Q1 2026 report says This is demand from looping, not yield. It was frozen by Aave after the exploit (see Dependencies).
- Sustainability: the documented base yield is staking and restaking rewards. The size of the Aave interest share is not quantified. Points are an extra incentive whose current value is not quantified in the sources.
- Historical yield ranges: Insufficient data. The files contain no rsETH APY series.
Dependencies & Integrations
rsETH depends on EigenLayer and the underlying staking tokens for backing, on Chainlink and token-specific oracles for prices, and on bridge providers for L2 copies.
- Backing assets: DefiLlama's dependency data shows stETH at 49.77% and ETH at 39.15% of backing, with ETHx (issued by Stader) at 11.06% .
- Restaking layer: EigenLayer, through
NodeDelegatorcontracts and four named node operators. - Price oracles:
LRTOraclereads per-asset oracles. The contracts list includes a Chainlink price oracle and oracles for sfrxETH, ETHx, swETH and rETH . L2 rate delivery uses LayerZero V1 today. A Chainlink-feed replacement with a 0.5% deviation threshold and 24-hour heartbeat is described as . - Bridges: LayerZero (verifiers include Canary, Horizen, LayerZero Labs and Nethermind) and Chainlink CCIP. DefiLlama's dependency graph assigns shares of 0.0503 to LayerZero v2 and 0.0357 to CCIP .
- Aave: Aave V3 is both a venue where rsETH is collateral and a place where Kelp earns interest on idle withdrawal ETH . On-chain,
LRTWithdrawalManagerreported an Aave balance of 281340748462778116397 raw units (about 281 ETH) . - Aave as collateral venue after the exploit: the service providers report That was the state on April 20, 2026. The current freeze status is not confirmed here.
- Chains: the docs list Ethereum, Arbitrum, Base, Linea, Ink and Mantle as supported . Bridge routing is contested (above). The on-chain read also records a chain-shutdown gap for Blast (chain id 81457) .
- Secondary markets: the measured rsETH/WETH depth on Ethereum is thin. DEX probes show a 0.04% price impact at $500,000 and a 26.83% impact at $2,000,000 . About 97.8% of that depth is on Fluid DEX.
TVL & Adoption
DefiLlama's latest captured snapshot shows $739.06M TVL, entirely on Ethereum, collected 2026-11-12 .
- Trend: TVL climbed from $100M at listing on 2023-12-21 to a monthly peak of $1.859B on 2025-02-01 .
- Around the exploit: DefiLlama's pre-incident TVL was $1.592B on 2026-04-17 . The monthly snapshot on 2026-06-01 was $1.315B .
- Recent decline: TVL was $859M on the 2026-07-01 snapshot. The path within June is not shown . Over the 30 days to 2026-11-12, DefiLlama's flow estimate is . Of that, about $152M is price effect and the remainder is net outflow of tokens.
| Month start | TVL (USD) | Source |
|---|---|---|
| 2024-03-01 | 627,462,531 | |
| 2024-06-01 | 1,159,848,618 | |
| 2025-10-01 | 1,820,453,307 | |
| 2026-04-01 | 1,299,744,384 | |
| 2026-11-01 | 860,038,926 |
- Chain distribution: 100% Ethereum in DefiLlama . L2 rsETH is a claim on lockbox assets, not separate TVL.
- Holder concentration (block 26395177, 2026-11-12): the largest holder is a proxy contract with 84.09% of rsETH supply . Its identity is not established in the file beyond its proxy label, so the exposure it represents is not confirmed. The LayerZero bridge adapter holds 4.99% and the CCIP lockbox 3.55% .
- Integrations: Kelp's September 2025 post claims 50+ DeFi protocols and 300,000+ restakers . This is a self-reported figure from before the exploit.
Technical Architecture
Core Ethereum contracts are upgradeable proxies behind a 10-day timelock. Several day-to-day controls, including the bridge adapter owner, sit with a 3-of-6 Safe with no delay. This section gives measured facts. The Governance report covers upgrade and pause control in detail.
Upgradability
- Pattern: transparent-style proxies with an EIP-1967 admin.
rsETH,LRTConfig,LRTDepositPoolandLRTOracleall report ProxyAdmin0xb61e0E39…Dc78. - Timelock: that ProxyAdmin is owned by
TimelockController0x49bD9989…35B1. Its minimum delay is 864000 seconds, which is 10 days, confirmed on-chain at block 26379467 . The docs also say 10 days . - Who proposes: on-chain, the External Admin Safe (6-of-11) is both proposer and canceller, and the executor role is open to anyone . The docs say proposers are the "respective admin multisigs" . For Ethereum, the on-chain read points to the 6-of-11 Safe, not the 3-of-6 Admin Safe.
- A second ProxyAdmin without a timelock:
0x7550eAEe…53fais owned directly by the Admin Safe (3-of-6) . Itsupgradefunction has a fastest delay of 0 seconds .- At block 26379467, two measured contracts name it as their proxy admin. One is labelled
EIGEN MerkleDistributor Programatic EIGEN(0x9bb6d4b9…eff1) . - The other is labelled
2. LRT Adaptor contract(0xb185d980…283a) . - The adaptor's contract name in the read is
GainAdapterETH, and it is not in the contracts table, so its link to Kelp's own product is unconfirmed. Other contracts using this ProxyAdmin were not measured.
- At block 26379467, two measured contracts name it as their proxy admin. One is labelled
- Recent upgrades: Kelp documents core upgrades on 2025-12-20 and 2026-01-24, each behind a 10-day timelock, with Bailsec audit reports linked . A Base pool upgrade was dated April 10, 2026 with a 3-day timelock.
Safes and thresholds (Ethereum, block 26379467)
| Safe | Threshold | Role per docs and on-chain reads |
|---|---|---|
External Admin 0xb3696a81…0a46d | 6 of 11 | DEFAULT_ADMIN_ROLE on LRTConfig; timelock proposer; CCIP lockbox owner |
Admin 0xb9577E83…0DA2 | 3 of 6 | Owns the second ProxyAdmin |
Manager 0xCbcdd778…29A1 | 3 of 6 | Day-to-day operations; owner of RSETH_OFTAdapter |
Pauser 0xb6AbB489…9559 | 1 of 5 | Pause only |
The docs give the same thresholds . On-chain reads confirm them at that block . The RSETH_OFTAdapter owner is the Manager Safe . Kelp's docs state that
Parameters that admins can change
- Manager Safe (3-of-6, no delay):
setMaxMintAmountPerDayandblockUserTransfers. The parsed privileges show a delay of 0 for each . - External Admin Safe (6-of-11, no delay):
unpause()andrecoverFrozenFunds. - Operator role: the Manager Safe and two externally owned accounts (single keys) hold
OPERATOR_ROLE. It gates fund-moving functions inLRTConverterandLRTWithdrawalManager. - Oracle and token controls added January 2026: per-address 24-hour transfer holds and fund recovery in the rsETH token . This gives Kelp the power to freeze addresses, as it used on September 15, 2026 (source reliability: unverified publisher).
Emergency mechanisms
Kelp's playbook has the Pauser Safe call LRTConfig.pauseAll() on Ethereum, then pause each L2. The Manager Safe then blocks transfers on lockboxes and attacker addresses . Unpausing needs an admin . All measured contracts reported paused: false at block 26379467 . A bug bounty offers up to $250,000 .
Bridge templates
| Route | Pattern | Funds sit | What Kelp sets / inherits |
|---|---|---|---|
| LayerZero OFT | Lock and release on Ethereum, mint on L2 | One shared lockbox, RSETH_OFTAdapter, for all LayerZero routes | Kelp sets the DVN set and thresholds, per route; LayerZero provides the message layer. The documented set is 4-of-4 with 64 confirmations |
| Chainlink CCIP | Lock and release (LockReleaseTokenPool 1.5.1) | One lockbox for CCIP | CCIP provides router and risk-management proxy; Kelp's Safe owns the pool and rate-limit admin |
| L2 deposit pools | Receive ETH on L2 and issue rsETH | Per-chain RSETHPool | Daily mint limit of 500 rsETH on Base, Linea, Ink and Mantle |
The 4-of-4 setting is a Kelp statement. The docs cite a live on-chain read and a Bailsec review of the verification script, but this report did not re-measure the DVN set.
- (external)
- (contract)
- (actor)
- (external)
- risk
- risk
- risk
Key Risks (Protocol-Level)
The main risk is that bridged rsETH depends on a lockbox that a forged message already emptied once. Kelp states mainnet backing is intact, and the Aave service providers (source reliability: community publication) describe mainnet rsETH as not directly affected.
- Where funds sit: deposits sit in
LRTDepositPool,NodeDelegatorcontracts and EigenLayer, plus lockboxes for bridged rsETH. The Aave report says - Bridge verification risk (realized): 116,500 rsETH (about $292M) left the lockbox through one forged message. Community accounts attribute this to a 1-of-1 verifier setting, which Kelp's post does not confirm. Assessment: the bridge is the most likely route for a repeat loss. Bridged copies are backed only by lockbox balances. The Safe that owns the adapter acts without a timelock. Kelp's 4-of-4 DVN change lowers single-verifier risk, but the routing conflict leaves exact exposure unclear.
- Shortfall on bridged rsETH: the April 20 report estimated $123.7M or $230.1M of Aave bad debt depending on how losses are shared . These were modelled scenarios. The later accounts above say the lockbox was refilled and the shortfall closed in full, which no official or independent source confirms. On-chain,
rsETHPriceequalshighestRsethPriceat block 26379467 . The two values match at that block only. They do not show whether the rate changed earlier. - Admin and key risk: a 3-of-6 Safe owns the bridge adapter and sets mint caps and fees with no delay. Two single-key accounts hold the operator role. Core upgrades are slower (10 days) but the executor role is open.
- Oracle risk: the rate comes from on-chain balances and oracle prices, rather than a DEX price. It relies on per-asset oracles (Chainlink and others) and, on L2s, a LayerZero-relayed rate. The Chainlink-based L2 replacement is still in audit per the docs.
- Restaking risk: exposure to EigenLayer slashing and to the four node operators. Kelp's own risk page lists smart contract risk and EigenLayer whale caps .
- Liquidity risk: instant withdrawal is "subject to availability of funds", and standard exit takes 15-21 days. DEX depth is thin at $2M trade size.
- Complexity: the on-chain read covers core contracts, two bridge stacks and multi-chain rate providers. Audit coverage is outside this report.
- Design assumptions that failed or are open: community accounts describe a single-verifier route that failed in April 2026. Cross-chain claims are assumed to stay fully backed. L2 rate relays are assumed fresh and in bounds.
Data Completeness
- Available: DefiLlama TVL snapshot, history and incident record. Official Kelp docs on restaking, security, oracle, cross-chain, timelocks and multisigs.
- Also available: on-chain control reads for Ethereum at block 26379467 (2026-11-09) and other chains. Incident accounts from Kelp, LayerZero, Aave service providers, CoinDesk, Chainalysis and Metrika.
- Source quality: most incident accounts are community publications or from unverified publishers, and the Kelp-versus-LayerZero responsibility dispute rests on them. Kelp's own post confirms the forged packet and the 116,500 rsETH release. The 1-of-1 verifier setting rests on community and unverified-publisher accounts and is not stated in Kelp's post.
- Missing or unresolved:
- An rsETH APY history.
- Official or independent confirmation that the lockbox was refilled and the bridged-rsETH shortfall closed in full. The accounts come from community and unverified publishers.
- The current DVN set measured on-chain. Kelp states 4-of-4, and this report did not re-measure it.
- Which routes are live today for the old and the new CCIP rsETH, since the Kelp docs page and the October 12 post differ. The post reports Optimism as one-way, back to Ethereum only.
- Whether contracts beyond the two measured ones use the non-timelocked ProxyAdmin, and what the
GainAdapterETHadaptor does for Kelp. - The identity of the proxy holding 84.09% of rsETH supply.
- Source code: no Kelp-specific claim here rests on a code review. Audited commits held in a private repository were not available to compare.
- Not assessed: Gain and Kernel (other products in the KernelDAO family), and TVL on other chains, which DefiLlama does not count for this record.
- Audits: audit coverage per component is outside this report's scope. The docs list audits by Sigma Prime, Code4rena, MixBytes and Bailsec between December 2023 and May 2026 .
146 citations · click markers to view evidence
Code Security
Code Security Analysis: Kelp
Kelp's core rsETH contracts have been reviewed repeatedly by independent firms, including MixBytes and Sigma Prime, but the collected sources do not show which audited revision the deployed contracts match. Auditors have not marked 8 high and 27 medium issue groups as fixed, and no critical group is unresolved. The multichain pool, bridge and administration contracts have no confirmed audit scope in the collected reports. Confidence is moderate, because deployed code was not compared with audited code.
Data Completeness
Based on available data: the findings ledger covers 15 published reports and write-ups; the SBSecurity analysis is an unverified-publisher publication (source reliability: unverified publisher), and the Rezolv write-ups and one repository copy are contest-finding copies rather than full audits. "Findings in ledger" below means critical, high and medium issue groups, counted once per group after cross-audit deduplication, with the latest audit status of each group; it is not a count of individual audit entries and not a count of confirmed live issues. Low and informational audit groups are excluded. The ledger holds 70 such groups. The ledger, the audit index and the coverage map were read in full, with only the specific audit sections quoted here; whole audits were not read.
Known gaps:
- Ten of the published reports sit on a host that is not verified as the auditor's own (for example kerneldao.com or a repository mirror), so their content carries lower source confidence. The Bailsec May 2026 report reviews only the off-chain verification script, not smart contracts.
- The audited commits of the Sigma Prime June 2025 and later reviews are not published in the collected repositories, so the audited code could not be compared with the deployed code.
- Collection gaps declared in the coverage map: three on-chain addresses on zkSync and Ethereum have no verified source; the repository Kelp-DAO/KelpDAO-contracts was not published; the audited code of ten reviewed revisions was not published; and the market-depth fact was not published.
- Inventory gaps: no current attributed address for NodeDelegator, GainLendingPool, GainAdapter and verify-onchain-config.
- Deployed-code checks used Sourcify runtime matches where available and, for several chains, Etherscan-listed source (an explorer claim, not a runtime match). On-chain facts were read at Ethereum block 26379467 (2026-11-09T19:36:23+00:00).
- External ratings are dated third-party snapshots, not proof of current security: the CertiK Skynet file lists a score of 78.8 and the Exponential file lists 7.0.
Audit Coverage
Published reviews of the Kelp contracts run from November 2023 to May 2026 and come from MixBytes, Sigma Prime, Bailsec/BailSec, a Code4rena contest and smaller write-ups. The auditor-hosted MixBytes report is dated March 2024. The Sigma Prime June 2024 report covers repository updates. The Code4rena report is a public contest report (source reliability: unverified publisher). The newest contract review is the BailSec differential review of December 2025; the May 2026 Bailsec report covers a script only.
| Auditor (as named in the report) | Tier | Date | Scope as stated | Publication note |
|---|---|---|---|---|
| MixBytes | tier 2 | March 2024 | Core contracts: LRTConfig, LRTOracle, LRTDepositPool, RSETH, NodeDelegator and utilities | published through a trusted auditor source |
| Sigma Prime (Liquid Restaking Token review) | tier 2 | December 2023 | LRTConfig, LRTDepositPool, LRTOracle, NodeDelegator and interfaces | published through a trusted auditor source |
| Sigma Prime (smart contract updates) | tier 2 | June 2024 | Whole repository, limited to a commit diff | published through a trusted auditor source |
| Sigma Prime (withdrawals) | tier 2 | June 2024 | LRTConverter, NodeDelegator, LRTDepositPool, LRTUnstakingVault, LRTWithdrawalManager and helpers | unverified publication host |
| Sigma Prime (PEPE upgrade integration) | tier 2 | September 2024 | The whole KelpDAO-contracts repository | unverified publication host |
| Sigma Prime (rsETH adapter) | tier 2 | November 2024 | Gain adapter and lending pool, ETH-derivative price oracles | unverified publication host |
| Sigma Prime (June 2025 upgrades) | tier 2 | June 2025 | FeeReceiver, NodeDelegatorHelper, L1Vault, L1VaultV2, RETHTokenWrapper, RSETH, LRTConfig | unverified publication host |
| Bailsec.io (RSETH) | tier not stated | September 2025 | not stated | unverified publication host |
| BailSec (Core, differential) | tier not stated | December 2025 | LRTConfig, LRTConverter, LRTDepositPool, LRTOracle, LRTWithdrawalManager, NodeDelegator, RSETH, LRTConfigRoleChecker | unverified publication host |
| Bailsec (verifier script review) | tier not stated | May 2026 | the off-chain verification script only; no smart contracts | unverified publication host |
| Code4rena contest report | not stated | December 2023 (contest of November 2023) | LRTConfig, LRTDepositPool, NodeDelegator, LRTOracle, RSETH, ChainlinkPriceOracle, LRTConfigRoleChecker | unverified publication host (source reliability: unverified publisher) |
| Code4rena contest findings, repository copy | not stated | November 2023 | same contest scope | unverified publication host (source reliability: unverified publisher); copy of a contest report |
| Rezolv (two contest finding write-ups) | tier not stated | November 2023 (one write-up); the other is undated | not stated | contest-finding copies, not full audits |
| SBSecurity analysis of ChainlinkPriceOracle and the rsETH overview | not stated | undated | ChainlinkPriceOracle | (source reliability: unverified publisher); not an audit |
Unconfirmed areas: for the multichain pool and bridge contracts (RSETHPoolV3 and variants, messengers, token wrappers, OFT contracts), the Safe and timelock contracts that administer them, and the cross-chain rate contracts, the coverage map shows no confirmed audit scope. That is a gap in what the collected reports state, not proof that no review exists. DefiLlama lists an April 2026 incident for Kelp (2026-04-18, 293,000,000 USD, technique "Cross-Chain Message Spoofing"). Kelp's own incident write-up states that a forged packet accepted by the bridge released 116,500 rsETH from the Ethereum bridge lockbox. The collected audits do not tie their scope to that bridge verification path.
Bug bounty and source verification: rsETH is covered by an Immunefi bounty of up to $250,000. The LRT deposit pool proxy is a Sourcify runtime match.
Audit findings by status
Counts cover critical, high and medium issue groups across all counted reports. These are protocol-wide: the counted audits cover LRT deposit, oracle, withdrawal, NodeDelegator, price-oracle, cross-chain pool and adapter code, so none of the counts belongs to a single product or chain.
| Severity | Total groups | Fixed per the auditor | Not marked fixed by the auditor |
|---|---|---|---|
| Critical | 3 | 3 | 0 |
| High | 22 | 14 | 8 |
| Medium | 45 | 18 | 27 |
The last column is an audit-status partition, not a count of confirmed live vulnerabilities.
The same groups split by reviewed scope, as total / fixed per the auditor / not marked fixed. The adapter scope is the November 2024 Sigma Prime review of the rsETH Adapter and Fractal's GainLendingPool. The script scope is the May 2026 Bailsec review of an off-chain configuration verifier, not of deployed contracts. Neither scope establishes fixes in the core rsETH contracts.
| Scope | Critical | High | Medium |
|---|---|---|---|
| rsETH core and related operations | 2 / 2 / 0 | 15 / 9 / 6 | 41 / 14 / 27 |
rsETH Adapter and GainLendingPool | 1 / 1 / 0 | 2 / 1 / 1 | 2 / 2 / 0 |
| Configuration verifier script | 0 / 0 / 0 | 5 / 4 / 1 | 2 / 2 / 0 |
Deployed code vs audits
Deployed code was not compared with the audited version: the map's line shares are 0.0% identical, 0.0% cosmetic, 0.0% changed, 100.0% unknown and 0.0% unaudited; the unknown bucket holds 175 files and 35610 lines because no audit states a confirmed scope revision (reason: scope not stated). Uncovered lines by reviewed criticality: critical 30732, high 2174, peripheral 2657, unrated 47, over the roles funds, access control, upgradeability, oracle, view utility and other. Files whose line counts are unavailable: 0. Code whose criticality is unassessed: 47 lines. Components not measured: 4. A zero identical share alone does not show that the code differs or lacks an audit; it shows that no comparison was possible.
Counts from the map: deployments known 128; with verified source 124; compared (with an audited revision or with every searched commit) 18; bound to a public commit 18; identical to an audited revision 0; with unaudited changes 0; still being searched 0; with an unresolved search 36. Source representation not comparable: 9 deployments. The map checks auditor-fixed findings against deployed files for 36 findings, and none can be determined because no fix review matched a deployed file; nothing is claimed from these. The binding of a deployment to a public commit names the public source, not an audit of it.
Collection gaps stated by the map: unverified source at three on-chain addresses; Kelp-DAO/KelpDAO-contracts not published; audited code of several reviewed commits not published; market depth not published. Uncollected scope is not evidence that published audits are missing. Inventory is partial; the map lists these inventory gaps: no attributed current address for NodeDelegator, GainLendingPool, GainAdapter or verify-onchain-config.
Rows below show gap and critical components first (30 table rows covering 43 of 74 listed items, including the 14 inventory names; 31 further components are not shown: 17 critical rows whose audit scope is not confirmed, many of them further Safe, OFT and wrapper deployments, 11 high-criticality oracle and rate rows and 3 peripheral OFT rows).
| Component | Deployment | Status | Audit and revision | Criticality | Note |
|---|---|---|---|---|---|
| LRTConverter | eip155:1:0x70daf8b0bfc846cc98b71d2f8ffdc91f4d2bbd51 | named in an audit scope; audited revision not confirmed | named in an audit scope; revision not confirmed | at least critical from assessed code (roles: funds, access control, upgradeability, oracle, view util, other) | version not compared; Sourcify runtime match; public source commit Kelp-DAO/LRT-rsETH@3dded88 (main) |
| LRTWithdrawalManager | eip155:1:0x0ecde3f414d1a245246d121e37191d9a63684e19 | named in an audit scope; audited revision not confirmed | named in an audit scope; revision not confirmed | at least critical from assessed code (roles: funds, access control, upgradeability, oracle, view util, other) | version not compared; Sourcify runtime match; search incomplete: some repository trees or repositories could not be read |
| ChainlinkPriceOracle | eip155:1:0xd73cd1aae045653474b873f3275ba2be2744c8b4 | named in an audit scope; more than one candidate audited revision | named in an audit scope; candidate audited revisions are ambiguous | at least critical from assessed code (roles: access control, upgradeability, oracle, view util) | version not compared; Sourcify runtime match; public source commit darrykane/888666-KelpDAO-contracts@1fdcd99 (main) |
| LRTConfig (LRT Config) | eip155:1:0xd4f475a7df199b3106f622a3a825ff399d4dafce | named in an audit scope; audited revision not confirmed | named in an audit scope; revision not confirmed | at least critical from assessed code (roles: funds, access control, upgradeability, view util) | version not compared; Sourcify runtime match; public source commit Kelp-DAO/LRT-rsETH@3dded88 (main) |
| LRTUnstakingVault | eip155:1:0x1fc8eebd7e1e61cc2cca005ee0f0d08417e5a2a4 | named in an audit scope; more than one candidate audited revision | named in an audit scope; candidate audited revisions are ambiguous | at least critical from assessed code (roles: funds, access control, upgradeability, view util) | version not compared; Sourcify runtime match; public source commit Kelp-DAO/LRT-rsETH@3dded88 (main) |
| RSETHRateProvider | eip155:1:0xf1cccba5558d31628216489a1435e068b1fd2c8a | named in an audit scope; more than one candidate audited revision | named in an audit scope; candidate audited revisions are ambiguous | at least critical from assessed code (roles: funds, access control, oracle, view util) | version not compared; Sourcify runtime match; public source commit Kelp-DAO/LRT-rsETH@630dc38 (tag:native-eth-deployment) |
| LRTDepositPool (LRT Deposit Pool) | eip155:1:0xea38dfa108318288f36f13d06e821a64acda8320 | named in an audit scope; audited revision not confirmed | named in an audit scope; revision not confirmed | at least critical from assessed code (roles: funds, access control, upgradeability, oracle, view util, other) | version not compared; Sourcify runtime match; public source commit Kelp-DAO/LRT-rsETH@3dded88 (main) |
| LRTOracle (LRT Oracle) | eip155:1:0xc59110239240761ccd3e670288443316e10dd271 | named in an audit scope; audited revision not confirmed | named in an audit scope; revision not confirmed | at least critical from assessed code (roles: funds, access control, oracle, view util) | version not compared; Sourcify runtime match; public source commit Kelp-DAO/LRT-rsETH@3dded88 (main) |
| RSETH (rsETH) | eip155:1:0x7159107483e623707c18c6e06cbc095bd0717783 | named in an audit scope; audited revision not confirmed | named in an audit scope; revision not confirmed | at least critical from assessed code (roles: funds, access control, upgradeability, view util) | version not compared; Sourcify runtime match; public source commit Kelp-DAO/LRT-rsETH@3dded88 (main) |
| zkSync | eip155:324:0x4f472e76ec322fdfa4011ae0d7cd8cabb947d270 | verified source not available | no audited revision identified | not assessed | version not compared; no verified source |
| Timelock Proposer | eip155:324:0xed38da849b20fa27b07d073053c5f5aae6a2db6b | verified source not available | no audited revision identified | critical (proposer for the timelock that owns the ProxyAdmin) | version not compared; no verified source |
| Scroll | eip155:534352:0xb6abb489aca4583833230f10b3a7670114d09559 | verified source not available | no audited revision identified | high (pause-only 1-of-5 Safe) | version not compared; no verified source |
| Timelock Proposer | eip155:534352:0xee68df9f661da6ed968ea4cbf7ec68fcfe375bc6 | verified source not available | no audited revision identified | critical (proposer for the timelock that owns the ProxyAdmin) | version not compared; no verified source |
| Audit-scope inventory names (14 entries: GainAdapter, GainLendingPool, HashStorage, IEigenStrategyManager, KelpDAO-contracts, METHPriceOracle, NodeDelegator, RETHTokenWrapper, RSETHPriceFeed, verify-onchain-config, WeETHPriceOracle, WstETHPriceOracle, plus 2 scope sentences from the Code4rena report) | none attributed | no current address attributed | named in audit scope lists | not assessed | version not compared; no deployed address is attributed to these names |
| GnosisSafeL2 (Timelock Proposer) | eip155:10:0xfb1bffc9d739b8d520daf37df666da4c687191ea; eip155:59144:0x3e5c63644e683549055b9be8653de26e0b4cd36e (+2 more) | audit scope not confirmed for this component | no audited revision identified | at least critical from assessed code (roles: funds, access control, upgradeability, view util, other) | version not compared; Sourcify runtime match; search incomplete: some repository trees or repositories could not be read |
| GnosisSafeProxy (Stable) | eip155:5000:0xb6abb489aca4583833230f10b3a7670114d09559; eip155:5000:0xfa3ba682f2210d05d087b331c71f5b34f16cd3e9 (+1 more) | audit scope not confirmed for this component | no audited revision identified | at least critical from assessed code (roles: upgradeability) | version not compared; Etherscan lists verified source (explorer claim); source representation not comparable |
| SafeL2 (BSC) | eip155:42161:0x29fcb43b46531bca003ddc8fcb67ffe91900c762; eip155:43114:0x29fcb43b46531bca003ddc8fcb67ffe91900c762 (+1 more) | audit scope not confirmed for this component | no audited revision identified | at least critical from assessed code (roles: funds, access control, upgradeability, view util, other) | version not compared; Sourcify runtime match; search incomplete: some repository trees or repositories could not be read |
| RsETHTokenWrapper | eip155:5000:0x6daf987d3486c65ff5bc1c5ae40fa50b6349c132; eip155:8453:0xcbb9b5faa23da272d2bdfc7bc3236b00631a8837 | audit scope not confirmed for this component | no audited revision identified | at least critical from assessed code (roles: funds, access control, view util) | version not compared; Etherscan lists verified source (explorer claim); public source commit Kelp-DAO/LRT-rsETH@3dded88 (main) |
| SafeProxy (Access Control & Multisigs) | eip155:59144:0xb6abb489aca4583833230f10b3a7670114d09559; eip155:8453:0xb6abb489aca4583833230f10b3a7670114d09559 | audit scope not confirmed for this component | no audited revision identified | at least critical from assessed code (roles: funds, upgradeability) | version not compared; Etherscan lists verified source (explorer claim) |
| ArbitrumMessenger | eip155:42161:0x58439f875cbfaa81b94cea89a934b6108bdec9f7 | audit scope not confirmed for this component | no audited revision identified | at least critical from assessed code (roles: funds) | version not compared; Sourcify runtime match; search incomplete: some repository trees or repositories could not be read |
| RSETHPoolV3 | eip155:5000:0xf1cccba5558d31628216489a1435e068b1fd2c8a | audit scope not confirmed for this component | no audited revision identified | at least critical from assessed code (roles: funds, access control, upgradeability, oracle, view util) | version not compared; Etherscan lists verified source (explorer claim); public source commit Kelp-DAO/LRT-rsETH@3dded88 (main) |
| AssetRegistry (Contract Deployments) | eip155:56:0x2fa25c6a9814f2e29837c3de0663c57b9fd8f88a | audit scope not confirmed for this component | no audited revision identified | at least critical from assessed code (roles: access control, upgradeability, view util, other) | version not compared; Sourcify runtime match; public source commit Kelp-DAO/kernel-smart-contracts-public@cf9a3bd (main) |
| L1Vault (L1Vault (Arbitrum)) | eip155:1:0x2f13d04b79d484b04358f64820cdb2725041e0e4 | audit scope not confirmed for this component | no audited revision identified | at least critical from assessed code (roles: funds, access control, oracle, view util) | version not compared; Sourcify runtime match; search incomplete: some repository trees or repositories could not be read |
| KERNEL_OFT (LayerZero KERNEL_OFT) | eip155:42161:0x6e401189c8a68d05562c9bab7f674f910821eacf | audit scope not confirmed for this component | no audited revision identified | at least critical from assessed code (roles: funds, access control, upgradeability, view util) | version not compared; Sourcify runtime match; source representation not comparable |
| RSETHPool | eip155:42161:0xa5e725d4a62e04dcae7fcb3bf8c888fd54d68fc1 | audit scope not confirmed for this component | no audited revision identified | at least critical from assessed code (roles: funds, access control, upgradeability, oracle, view util) | version not compared; Sourcify runtime match; search incomplete: some repository trees or repositories could not be read |
| L1VaultV2 (L1Vault (Linea)) | eip155:1:0xb0313a93f90f54157f329dc647cb42cc763b5974 | audit scope not confirmed for this component | no audited revision identified | at least critical from assessed code (roles: funds, access control, upgradeability, oracle, view util, other) | version not compared; Sourcify runtime match; search incomplete: some repository trees or repositories could not be read |
| KernelConfig (Contract Deployments) | eip155:56:0x1ecfdc02fc51d5c5499ee97a2cdc5ee3db084eaa | audit scope not confirmed for this component | no audited revision identified | at least critical from assessed code (roles: access control, upgradeability, view util, other) | version not compared; Sourcify runtime match; search incomplete: some repository trees or repositories could not be read |
| AGETHPoolV3 | eip155:59144:0x32f952b0f16de076996373f8693f6f2eb8a15d5b | audit scope not confirmed for this component | no audited revision identified | at least critical from assessed code (roles: funds, access control, upgradeability, oracle, view util) | version not compared; Sourcify runtime match; search incomplete: some repository trees or repositories could not be read |
| LineaMessenger | eip155:59144:0x838686d23521435b528f68cde6404c07ae007299 | audit scope not confirmed for this component | no audited revision identified | at least critical from assessed code (roles: funds, access control, oracle, view util) | version not compared; Sourcify runtime match; search incomplete: some repository trees or repositories could not be read |
| FactoryBurnMintERC20 (CCIP (Chainlink) RSETH (New)) | eip155:59144:0xb999ea589e0a1cce9153601dac2d6e203c2fd577 | audit scope not confirmed for this component | no audited revision identified | at least critical from assessed code (roles: funds, access control, view util) | version not compared; Sourcify runtime match; source representation not comparable |
Not-checked rows share one cause: no included audit states which revision of the code it reviewed, so the link from audit scope to deployed code is not shown. That is not a mismatch.
Findings by Severity
Findings below come from the ledger, one entry per issue group; fixed groups appear only as counts. "Check" is our comparison of the code with what the finding describes, from verified deployed source where available, otherwise from the repository default branch (Kelp-DAO/LRT-rsETH@3dded88, pushed 2026-02-10, not archived). A visible fix in the repository does not prove deployment, and a Check never changes an audit status.
Critical
Total groups: 3; fixed per the auditor: 3; not marked fixed: 0. No group in this severity is marked other than fixed by the auditors, so there is nothing to check.
High
Total groups: 22; fixed per the auditor: 14; not marked fixed: 8.
- Possible arbitrage from Chainlink price discrepancy (Code4rena contest findings report, November 2023; Code4rena contest, December 2023) (source reliability: unverified publisher) — — Audit status: disputed (latest status from Code4rena contest, December 2023) — Check: the checked version of the code does not contain the described code in this form (deployed-code check)
- Loss of value upon mint via direct donation (share price manipulation) / The price of rsETH could be manipulated by the first staker (Code4rena contest findings report, November 2023; Code4rena contest, December 2023) (source reliability: unverified publisher) — — Audit status: acknowledged (latest status from Code4rena contest, December 2023) (earlier members: not stated in Code4rena contest findings report, November 2023) — Check: the checked version of the code does not contain the described code in this form (deployed-code check)
- The reward calculation may be blocked until contract upgrade (MixBytes, March 2024) — — Audit status: acknowledged (latest status from MixBytes, March 2024) — Check: not marked fixed per the auditor (acknowledged); our code check is unavailable: the audited repository code was not available
- Potential yield stealing due to centralized oracle updates / Stale RSETH Price Referenced In Core Operations / Stale price enables yield extraction attack (Bailsec.io, September 2025; MixBytes, March 2024; Sigma Prime, June 2025) — — Audit status: acknowledged (latest status from Bailsec.io, September 2025) — Check: the code this finding describes is unchanged in the source verified for eip155:1:0xea38dfa108318288f36f13d06e821a64acda8320 (Sourcify runtime match) (identical to public commit Kelp-DAO/LRT-rsETH@3dded88) (deployed-code check)
- Price Changes Result In Stuck Tokens (Sigma Prime, November 2024) — — Audit status: acknowledged (latest status from Sigma Prime, November 2024) — Check: not marked fixed per the auditor (acknowledged); our code check is unavailable: the audited code was not found in the collected files
- A whale can entirely brick the profit logging for rsETH for the entire day / Daily fee limit blocks price updates (Bailsec.io, September 2025) — — Audit status: acknowledged (latest status from Bailsec.io, September 2025) — Check: the code this finding describes is unchanged in the source verified for eip155:1:0xc59110239240761ccd3e670288443316e10dd271 (Sourcify runtime match) (identical to public commit Kelp-DAO/LRT-rsETH@3dded88) (deployed-code check)
- Hub-and-spoke topology is assumed by iteration order, not verified (Bailsec, May 2026) — — Scope note: this finding concerns the off-chain verification script, not a deployed contract — Audit status: partially fixed (latest status from Bailsec, May 2026) — Check: not marked fixed per the auditor (partially fixed); our code check is unavailable: the audited code was not found in the collected files
- First depositor can brick the protocol via rounding in getRsETHAmountToMint (analysis by SBSecurity (source reliability: unverified publisher), not an audit) — — Audit status: open (latest status from analysis by SBSecurity (source reliability: unverified publisher), not an audit) — Check: the checked version of the code does not contain the described code in this form (deployed-code check)
Medium
Total groups: 45; fixed per the auditor: 18; not marked fixed: 27.
- After changing the strategy of particular asset, users deposits will become permanently lost/locked in the Strategy. Additionally getAssetDistributionData() gets the latest strategy that has the asset balance deposited / Strategy update without migration of user deposits breaks rsETH share price accounting / Update in strategy will cause wrong issuance of shares (Code4rena contest findings report, November 2023; Code4rena contest, December 2023; Rezolv) (source reliability: unverified publisher) — — Audit status: disputed (latest status from Code4rena contest, December 2023) (earlier members: not stated in Rezolv) — Check: the fix is visible in the source verified for eip155:1:0xd4f475a7df199b3106f622a3a825ff399d4dafce (Sourcify runtime match) (identical to public commit Kelp-DAO/LRT-rsETH@3dded88) (deployed-code check)
- LRTDepositPool.sol#depositAsset() lack of min minted amount received and expiration timestamp checks / Lack of slippage control on LRTDepositPool.depositAsset (Code4rena contest findings report, November 2023; Code4rena contest, December 2023; Rezolv) (source reliability: unverified publisher) — — Audit status: not stated (latest status from Code4rena contest, December 2023) (earlier members: acknowledged in Code4rena contest findings report, November 2023) — Check: the fix is visible in the source verified for eip155:1:0xea38dfa108318288f36f13d06e821a64acda8320 (Sourcify runtime match) (identical to public commit Kelp-DAO/LRT-rsETH@3dded88) (deployed-code check)
- Skewed rsETH price/mint calculations if future LSTs use decimals other than 1e18 (Code4rena contest findings report, November 2023) (source reliability: unverified publisher) — — Audit status: not stated (latest status from Code4rena contest findings report, November 2023) — Check: the code this finding describes is unchanged in the source verified for eip155:1:0xc59110239240761ccd3e670288443316e10dd271 (Sourcify runtime match) (identical to public commit Kelp-DAO/LRT-rsETH@3dded88) (deployed-code check)
- Potential arbitrage opportunity in the rsETH price calculation for the unpriveleged users (MixBytes, March 2024) — — Audit status: acknowledged (latest status from MixBytes, March 2024) — Check: the checked version of the code does not contain the described code in this form (deployed-code check)
- Unsafe contract deployment and update process (MixBytes, March 2024) — — Audit status: acknowledged (latest status from MixBytes, March 2024) — Check: not marked fixed per the auditor (acknowledged); our code check is unavailable: the audited repository code was not available
- Protocol functionality lockout upon adding a new token (MixBytes, March 2024) — — Audit status: acknowledged (latest status from MixBytes, March 2024) — Check: the code this finding describes is unchanged in the source verified for eip155:1:0xd4f475a7df199b3106f622a3a825ff399d4dafce (Sourcify runtime match) (identical to public commit Kelp-DAO/LRT-rsETH@3dded88) (deployed-code check)
- Centralization risks (MixBytes, March 2024) — — Audit status: acknowledged (latest status from MixBytes, March 2024) — Check: the checked version of the code does not contain the described code in this form (deployed-code check)
- The possible inability to initialize the eigenPod field (MixBytes, March 2024) — — Audit status: acknowledged (latest status from MixBytes, March 2024) — Check: not marked fixed per the auditor (acknowledged); our code check is unavailable: the audited repository code was not available
- The Ether can't be withdrawn from the NodeDelegator contract under certain conditions (MixBytes, March 2024) — — Audit status: acknowledged (latest status from MixBytes, March 2024) — Check: the fix is visible in the source verified for eip155:1:0xea38dfa108318288f36f13d06e821a64acda8320 (Sourcify runtime match) (identical to public commit Kelp-DAO/LRT-rsETH@3dded88) (deployed-code check)
- NDC Index Shuffling Issue In LRTDepositPool / transferAssetToNodeDelegator may be frontrun in order to force funds to be sent to incorrect delegator (Bailsec.io, September 2025; Sigma Prime, June 2024) — — Audit status: acknowledged (latest status from Bailsec.io, September 2025) — Check: the code this finding describes is unchanged in the source verified for eip155:1:0xea38dfa108318288f36f13d06e821a64acda8320 (Sourcify runtime match) (identical to public commit Kelp-DAO/LRT-rsETH@3dded88) (deployed-code check)
- FeeReceiver Balance Is Not Included In TVL (Sigma Prime, September 2024) — — Audit status: acknowledged (latest status from Sigma Prime, September 2024) — Check: the code this finding describes is unchanged in the source verified for eip155:1:0xea38dfa108318288f36f13d06e821a64acda8320 (Sourcify runtime match) (identical to public commit Kelp-DAO/LRT-rsETH@3dded88) (deployed-code check)
- Incorrect stakedButUnverifiedNativeETH Accounting Can Overinflate TVL (Sigma Prime, September 2024) — — Audit status: acknowledged (latest status from Sigma Prime, September 2024) — Check: the default-branch snapshot (Kelp-DAO/LRT-rsETH@3dded88) does not contain the described code in this form; deployment not confirmed (repository check)
- getTotalAssetDeposits returns incorrect value for ETH (Bailsec.io, September 2025) — — Audit status: acknowledged (latest status from Bailsec.io, September 2025) — Check: the checked version of the code does not contain the described code in this form (deployed-code check)
- stETH 1-2 wei corner case breaks composability across codebase (Bailsec.io, September 2025) — — Audit status: acknowledged (latest status from Bailsec.io, September 2025) — Check: the code this finding describes is unchanged in the source verified for eip155:1:0xea38dfa108318288f36f13d06e821a64acda8320 (Sourcify runtime match) (identical to public commit Kelp-DAO/LRT-rsETH@3dded88) (deployed-code check)
- Missing staleness verification in getAssetPrice (Bailsec.io, September 2025) — — Audit status: acknowledged (latest status from Bailsec.io, September 2025) — Check: the code this finding describes is unchanged in the source verified for eip155:1:0xd73cd1aae045653474b873f3275ba2be2744c8b4 (Sourcify runtime match) (identical to public commit darrykane/888666-KelpDAO-contracts@1fdcd99) (deployed-code check)
- Storage collision in LRTOracle upgrade (Bailsec.io, September 2025) — — Audit status: acknowledged (latest status from Bailsec.io, September 2025) — Check: the checked version of the code does not contain the described code in this form (deployed-code check)
- sweepRemainingAssets may be DOSed indefinitely (Bailsec.io, September 2025) — — Audit status: acknowledged (latest status from Bailsec.io, September 2025) — Check: the code this finding describes is unchanged in the source verified for eip155:1:0x0ecde3f414d1a245246d121e37191d9a63684e19 (Sourcify runtime match) (deployed-code check)
- User can hedge his losses against slashing with an instant withdraw (Bailsec.io, September 2025) — — Audit status: acknowledged (latest status from Bailsec.io, September 2025) — Check: the code this finding describes is unchanged in the source verified for eip155:1:0x0ecde3f414d1a245246d121e37191d9a63684e19 (Sourcify runtime match) (deployed-code check)
- Instant withdrawals can brick withdrawal queue (Bailsec.io, September 2025) — — Audit status: acknowledged (latest status from Bailsec.io, September 2025) — Check: the checked version of the code does not contain the described code in this form (deployed-code check)
- RsETH is not compatible with CCIP self service registration (Bailsec.io, September 2025) — — Audit status: acknowledged (latest status from Bailsec.io, September 2025) — Check: the checked version of the code does not contain the described code in this form (deployed-code check)
- maxAmountToDepositBridgerAsset only accounts for one altRsETH token (Bailsec.io, September 2025) — — Audit status: acknowledged (latest status from Bailsec.io, September 2025) — Check: the code this finding describes is unchanged in the source verified for eip155:5000:0x6daf987d3486c65ff5bc1c5ae40fa50b6349c132 (Etherscan verified source, explorer claim) (identical to public commit Kelp-DAO/LRT-rsETH@3dded88) (deployed-code check)
- LRTManager can pause minting by setting maxMintAmountPerDay to 0 (BailSec, December 2025) — — Severity: Medium (estimated); this is an AI estimate because the audit states no rating, and the reason is: the auditor labeled this 'Governance', a non-severity label; a compromised LRTManager can halt rsETH minting via maxMintAmountPerDay = 0, a denial-of-service impact. — Audit status: acknowledged (latest status from BailSec, December 2025) — Check: the code this finding describes is unchanged in the source verified for eip155:1:0x7159107483e623707c18c6e06cbc095bd0717783 (Sourcify runtime match) (identical to public commit Kelp-DAO/LRT-rsETH@3dded88) (deployed-code check)
- ChainlinkPriceOracle uses stale latestAnswer without safety checks (analysis by SBSecurity (source reliability: unverified publisher), not an audit) — — Audit status: open (latest status from analysis by SBSecurity (source reliability: unverified publisher), not an audit) — Check: not marked fixed per the auditor (open); our check row for this group is treated as not checked, because the finding names
latestAnswerwhile the verified deployed source callslatestRoundData, so an unchanged-code verdict is not supported. The verified source reads only the price answer fromlatestRoundDataand shows no freshness check in that excerpt (deployed-code evidence). - No way to remove deprecated supported assets from LRTConfig (analysis by SBSecurity (source reliability: unverified publisher), not an audit) — — Audit status: open (latest status from analysis by SBSecurity (source reliability: unverified publisher), not an audit) — Check: the fix is visible in the source verified for eip155:1:0xd4f475a7df199b3106f622a3a825ff399d4dafce (Sourcify runtime match) (identical to public commit Kelp-DAO/LRT-rsETH@3dded88) (deployed-code check)
- stETH rebase can hit deposit limits and DoS deposits (analysis by SBSecurity (source reliability: unverified publisher), not an audit) — — Audit status: open (latest status from analysis by SBSecurity (source reliability: unverified publisher), not an audit) — Check: the checked version of the code does not contain the described code in this form (deployed-code check)
- No withdrawal functionality for users or from EigenLayer (analysis by SBSecurity (source reliability: unverified publisher), not an audit) — — Audit status: open (latest status from analysis by SBSecurity (source reliability: unverified publisher), not an audit) — Check: the default-branch snapshot (Kelp-DAO/LRT-rsETH@3dded88) does not contain the described code in this form; deployment not confirmed (repository check)
- High centralization risk from LRTManager and LRTAdmin roles (analysis by SBSecurity (source reliability: unverified publisher), not an audit) — — Audit status: open (latest status from analysis by SBSecurity (source reliability: unverified publisher), not an audit) — Check: the code this finding describes is unchanged in the source verified for eip155:1:0x7159107483e623707c18c6e06cbc095bd0717783 (Sourcify runtime match) (identical to public commit Kelp-DAO/LRT-rsETH@3dded88) (deployed-code check)
Recurring Patterns
- Price and rate handling recurs across audits. The stale-price and yield-extraction group was raised by MixBytes, Sigma Prime and Bailsec.io and is acknowledged; our check found the described code unchanged in the verified deposit pool source. Oracle staleness, rate-update fee limits and rsETH price arbitrage groups also remain not marked fixed.
- Accounting and deposit-flow issues recur: the first-staker and share-price groups (contest reports and the SBSecurity analysis), the NodeDelegator index and frontrunning group (Sigma Prime and Bailsec.io), and FeeReceiver and unverified native ETH accounting groups (Sigma Prime) are all acknowledged or disputed rather than fixed.
- Privileged-role risk recurs: the centralization groups raised by MixBytes and by the SBSecurity analysis, and the group in which LRTManager can set the daily mint limit to zero. The last is an estimated severity. Our check found the described code unchanged in the verified source for the rsETH token contract.
- Strategy-change and asset-listing issues were reported by the Code4rena contest, Rezolv and a repository copy; they remain disputed or not stated rather than fixed.
- Fixed groups (3 critical, 14 high, 18 medium) are counted only; no pattern is claimed from them.
Code Quality Assessment
- Review depth is substantial: several independent firms (MixBytes, Sigma Prime, Bailsec) and a Code4rena contest (source reliability: unverified publisher) reviewed the core deposit, oracle and withdrawal contracts between 2023 and 2026. The fix rate per the auditors is 3 of 3 critical, 14 of 22 high and 18 of 45 medium groups; the remaining groups are mostly acknowledged by the team rather than changed.
- Coverage is the weak point: no deployed contract is shown identical to an audited revision, 60 listed components mostly have no confirmed audit scope or revision, and the audited commits of recent reviews are not public. Most bridge, pool and Safe contracts have no confirmed audit scope.
- Source is public for the core contracts (Sourcify runtime matches for the main Ethereum contracts), which helps outside review.
- Known compiler bugs: the verified Kelp core contracts in the collected sources use Solidity 0.8.27 (deposit pool, oracle, config, withdrawal, unstaking vault, converter, fee receiver and token), 0.8.21 (ChainlinkPriceOracle) and 0.8.22 (OFT contracts). For these versions the bug list names one medium bug, SOL-2026-2, fixed in 0.8.36 and conditional on the viaIR pipeline: "Local variables of a function involved in mutual recursion may spuriously be moved to fixed memory offsets and overwritten across recursive calls." Link: https://blog.soliditylang.org/2026/07/09/unsound-spill-in-mutual-recursion-bug/. Five further bugs rated low, very low or low/medium are listed for each of those versions. Whether any condition applies was not checked, so this is not a finding.
Score: 5.0/10
Rubric security-score-v2.6; data completeness 1.00
- Audit coverage
- 5.50 points
- Audit depth
- 9.50 points
- Unresolved findings
- 0.22 points
- Assurance
- 7.50 points
- High · open
- 1 findings
- High · partially fixed
- 1 findings
- High · acknowledged
- 5 findings
- High · disputed
- 1 findings
- Medium · open
- 5 findings
- Medium · not stated
- 2 findings
- Medium · acknowledged
- 19 findings
- Medium · disputed
- 1 findings
| Severity | Groups | Fix visible (deployed code) | Fix visible (repository only) | Described code still there (deployed code) | Described code still there (repository only) | Not checked or inconclusive |
|---|---|---|---|---|---|---|
| Critical | 0 | 0 | 0 | 0 | 0 | 0 |
| High | 8 (1 not scored) | 0 | 0 | 2 | 0 | 6 (1 not scored) |
| Medium | 27 (4 not scored) | 4 (1 not scored) | 0 | 11 (2 not scored) (2 low confidence or archived) | 0 | 12 (1 not scored) |
Not scored: disputed groups and groups excluded for a later audit fix or privileged role.
Low confidence or archived: the check was not confident enough, or the repository is archived; the described code may still be there.
Deployed code: the source verified for the live contract. Repository only: the protocol's public repository; deployment not confirmed. “Described code still there” is a limited code signal, not a confirmed vulnerability: no exploit was tested; other contracts, configuration or operations may mitigate it. Findings the auditor marked fixed are trusted, not re-checked. Ask the team whether these are fixed or mitigated. Raw penalties: deployed code 6.36, repository only 0.00, audit status alone 3.42; Unresolved findings loses min(6.36 + 0.00 + 3.42, 10) of 10 points.
| Factor | Level | Evidence | Weight | Points |
|---|---|---|---|---|
| Audit coverage (computed) | L2 | supported | 0.35 | 5.50 |
| Audit depth | L4 | supported | 0.20 | 9.50 |
| Unresolved findings (computed) | L0 | supported | 0.30 | 0.22 |
| Assurance | L3 | supported | 0.15 | 7.50 |
66 citations · click markers to view evidence
Economic Security
Economic Security Analysis: Kelp
Kelp's rsETH rate and backing are computed on-chain from held balances, which is a sound base, but the token is thinly traded and almost entirely parked in one lending-market contract. Its principal also rests on EigenLayer, Lido stETH, Stader ETHx and cross-chain bridges, and one of those bridges failed in April 2026 for about USD 293 million. Assessment: the on-chain design is stronger than the market and dependency structure around it. The April recovery is described as complete only by weak sources, and no Kelp-owned loss fund is documented. TVL is USD 739 million, all on Ethereum, and it has shrunk sharply in 30 days.
TVL & Liquidity Profile
Kelp holds USD 739.06 million of TVL, 100% on Ethereum, and the Ethereum rsETH market is thin (about USD 0.5 million of depth at 2% price impact) against USD 737 million of supply.
- TVL is USD 739,058,507.82 and the only chain is Ethereum. The category is Liquid Restaking. The DefiLlama snapshot was collected 2026-11-12T20:47Z, so the figures below are same-day.
- The 7-day change is -12.76%, from the same 2026-11-12 snapshot. The 1-day change is not used because one day of TVL movement says little about structure.
- Composition at 2026-11-12: stETH 49.78% (USD 367.87 million), ETH 39.16% (USD 289.38 million) and ETHx 11.07% (USD 81.81 million).
- stETH
- 367.87 USD m
- ETH
- 289.38 USD m
- ETHx
- 81.81 USD m
Outflows. Over 30 days TVL fell by USD 240.4 million, of which USD 151.9 million is price effect and USD 88.5 million is token-quantity outflow valued at current prices. Over 7 days the figures are USD -153.6 million total, USD -139.8 million price effect and USD -13.9 million quantity change. Over 90 days the price effect is positive (USD +77.4 million) and the quantity change is USD -187.2 million. Assessment: most of the 7-day drop is price movement, while the 90-day token outflow is large. No workspace source gives the reason for the outflow.
Measured market depth. The Ethereum rsETH probe records USD 500,000 of depth at 2% impact and at 50 bps. The 50 bps figure is a lower bound from coarse probes (USD 0.1M, 0.5M, 2M), not an exact capacity. The USD 100,000 probe moves price 0.03%, the USD 500,000 probe 0.04% and the USD 2 million probe 26.83%. Venues are Fluid DEX (97.81%) and Uniswap v4 (2.19%). USD 0.5 million is under 0.1% of supply and below the USD 2 million bar the rubric uses for L2. Other-chain copies show no measured depth: the Arbitrum rsETH, Base wrsETH and Linea wrsETH rows read zero and Mantle has no quote source. Those are read as thin or unmeasured markets, not proof of no liquidity.
Native exit queue. Withdrawals exist only on Ethereum and pay out as ETH, ETHx or stETH. Instant withdrawal takes the assets immediately for a 0.5% fee, and standard withdrawal is free but takes 15-21 days. The instant route is documented as under 2 minutes but "Subject to availability of funds", so its capacity is not guaranteed. The standard queue is a stated bound of 21 days, and it is the only route that does not depend on the thin market or on that availability caveat.
Mint caps. Ethereum rsETH has an on-chain daily mint cap of 5,000 rsETH, and the Base, Linea, Ink and Mantle pools are capped at 500 rsETH/day each. The on-chain read at block 26379467 (2026-11-09) shows the 5,000 rsETH cap set. At the 2026-11-12 price of USD 2,180.87 that is about USD 10.9 million of new rsETH per day on Ethereum and about USD 1.1 million per day on each capped L2 pool (my calculation). These caps limit new minting. They do not cap redemption or the release of rsETH from the bridge lockbox. The on-chain read of the rsETH LayerZero adapter route to Arbitrum records no rate limit, while the CCIP lock-release pool has token buckets (see Economic Attack Vectors).
Holder concentration (Ethereum rsETH). At block 26395177 (as of 2026-11-12), the top holder owns 84.09% of the 340,580.55 rsETH supply, the top 10 own 93.16% and protocol-owned contracts own 5.17%. The top holder is a contract labelled as an Aave-style proxy. Assessment: this is most likely the Aave V3 Core rsETH aToken, because the Aave Core rsETH reserve shows 288,884.98 rsETH supplied, close to the holder's balance. No source states that identification, so treat it as an inference. If it is right, the 84% is many depositors' collateral in one lending market, and its exit depends on that market. The Aave rsETH reserve is currently frozen at the 2026-11-09 read.
Oracle Risk
rsETH's rate is computed on-chain from held balances with a 1% per-update bound, which makes it hard to manipulate in a single block, but the rate can lag, an acknowledged audit finding describes arbitrage on a stale rate, and the L2 rate is relayed over a messaging layer.
- What is priced. rsETH is a reward-bearing token: its ETH value rises as rewards accrue, so the rate is share accounting, not a market price. The
LRTOraclereads every supported asset balance, prices each, sums to a total ETH value and divides by rsETH supply. The docs state that the rate cannot be moved by a single swap or flash-loan sandwich in one block. - Independent cross-check.
TotalETHValueReader.getLiveRsETHPrice()recomputes the price view-only and serves as proof of reserves. At block 26379467 the oracle reads 1.084187 ETH per rsETH and the live reader reads 1.084227 ETH, a gap of about 0.004% (my calculation). The oracle read is three days old at the research date. The two readings are a snapshot comparison from Kelp's own contracts and do not establish that backing is adequate. - Bounds. A single update is capped by
pricePercentageLimit, documented as 1%. The on-chain value is 10000000000000000. Conditional on the contract using 1e18 precision, as the repository code comment says, that equals 1%. The docs also state the oracle keeps an all-time-high price and auto-pauses deposits, withdrawals and oracle updates on a downside breach. - Who controls it. The 6-of-11 admin multisig alone changes the limit and its bounds; the 3-of-6 manager multisig submits routine price updates. The oracle read shows it unpaused.
- Per-asset inputs. The deposit assets are priced through per-asset oracles (Chainlink and others) listed in the contracts table. The 11.07% ETHx part is priced through an
EthXPriceOraclethat reads Stader's stake pools manager. A dependency report on Stader (background context only) describes the Stader exchange rate as pushed through a three-node oracle bounded byerChangeLimit(15 of 10,000 basis points, loosenable by a 3-of-6 Safe with zero delay). - L2 rate path. On L2s the rate is relayed: the L1
RSETHMultiChainRateProviderbroadcasts it as a LayerZero V1 message toRSETHRateReceivercontracts. A Chainlink rate feed (0.5% deviation, 24-hour heartbeat) with staleness assertions is planned and, per the docs, in audit. No cited source describes a staleness check on the current L2 consumer, and L2 TVL is a small share of the total, so this is a smaller exposure than the Ethereum path. - Audit findings on rate freshness. An audit of the withdrawal manager rates a stale-price issue High and records it as acknowledged: an attacker deposits at a stale low price, forces a price update, then withdraws instantly at the new price, which pays when the share-price rise exceeds the 0.5% instant fee. A second finding of the same audit says the oracle reverts when the daily fee-mint limit is reached, which leaves stale prices in place; it is also High and acknowledged. The audit appendix notes that a manager-only update function can bypass some price-threshold restrictions. These audit files have unverified publication provenance (source reliability: unverified publisher), so they are read as lower-confidence evidence; the protocol is not marked down for the publication host alone.
- Downstream pricing. The April 2026 loss did not come from the rate oracle. It came from a forged bridge message. The lending market then accepted the unbacked token as collateral (see Economic Attack Vectors).
Flash Loan & MEV Exposure
The core rate is not exposed to in-block manipulation, and the cited exposure sits in lending markets that take rsETH as collateral and in the instant-withdrawal fee.
- The rsETH rate is computed from balances and the docs say a flash-loan sandwich inside one block cannot move it.
updateRSETHPrice()is public and unpaused-only in the repository code, so anyone can trigger an update within the 1% bound. The call path for timing games around the update is not discussed in the sources, beyond the audit finding on stale prices in the Oracle Risk section. - The DEX pools hold little depth (USD 0.5 million at 2%), so a market-price view of rsETH is easy to move. Any lending market that prices rsETH from the market rather than from the protocol rate would be exposed to that. Aave's rsETH price source is not read in the workspace, so this is a gap.
- Instant withdrawal costs 0.5%, which prices out a cheap round trip between market and protocol. Assessment: that fee limits arbitrage of small price moves, but the audit finding above shows it does not stop a stale-rate round trip when the rate step exceeds 0.5%; the 15-21 day standard queue is the fee-free route.
- Mint caps bound abuse of the deposit side, as described in the liquidity section.
Tokenomics Risk: Yield, Leverage & Reflexivity
Returns come from named ETH staking and restaking rewards, the protocol's own token KERNEL does not back rsETH in any cited source, and leverage exists in user loops on Aave with a cited bound but no cited unwind rule.
- Yield source. Rewards are not paid out as extra tokens. The ETH each rsETH represents rises as staking and restaking rewards accrue to the assets underneath. No workspace source splits yield between revenue and token emissions, so that split is not cited.
- Fees. DefiLlama records protocol fees of USD 2.08 million over 30 days and USD 34.59 million over 365 days, with revenue of USD 0.10 million and USD 1.98 million. These are protocol-level figures, not the APY a user receives. A July 2024 risk-provider review says Kelp takes a 10% fee on ETH staking rewards for direct ETH deposits and none on LST deposits (source reliability: established; the review is older than the other facts here). The repository caps the configurable protocol fee at 1500 basis points, on the condition that the deployed contract uses a 10,000 denominator as the repository's oracle code does. The fee is taken by minting rsETH, capped on-chain at 20 rsETH per day.
- Own-token reflexivity. The composition of TVL is stETH, ETH and ETHx only. Neither KERNEL nor rsETH itself is in the measured composition, so no own-token backing is evident. KERNEL has no measured DEX depth, and its holder data shows the top 10 holders own 90.47% of the supply. That matters only if KERNEL later backs claims; no source says it does.
- Leverage. Kelp's own contracts add no leverage in the cited documents. Users can loop rsETH against ETH on Aave: the E-Mode went live in January, with up to 93% LTV and about 14x leverage on rsETH/ETH loops (source reliability: official protocol blog). No cited source gives an unwind rule for those loops; liquidation is handled by the lending market, outside Kelp's contracts.
- What happened in April. An attacker deposited stolen rsETH into Aave and borrowed WETH, which pushed WETH utilization to 100% and led Aave to freeze the reserves. The loss started with the forged bridge message, not with forced deleveraging of looping users, so it is not treated as a yield-design failure. (source reliability: established publisher)
- Own-token issues not scored here. Holder concentration of rsETH and KERNEL is covered under liquidity; governance of the 6-of-11 safe belongs to the governance report.
Solvency, Liquidation & Bad Debt
Kelp does not run a debt book, its backing is verifiable on-chain and exits are open on Ethereum, but the pause power has no stated maximum duration, the April 2026 remote-chain shortfall is described as refilled only by weak sources, and no Kelp-owned loss fund is documented.
- Backing. Backing is the on-chain balance of deposit assets, summed by the proof-of-reserves reader. At block 26379467 the reader reports a total ETH value in the protocol of about 374,194 ETH against 345,124 rsETH supply. That is 1.0842 ETH per rsETH (my calculation), matching the oracle. No independent attestation is cited, so the check rests on Kelp's own reader contract.
- Exit status. At the 2026-11-09 read (three days old) the oracle, deposit pool and withdrawal manager are not paused. The withdrawal manager reads a raw instant-withdrawal fee of 50 and zero withdrawal delay blocks on the instant route; the repository code defines the fee in basis points (1 = 0.01%), so 50 is 0.5% only if the deployed contract uses that unit, and the documentation separately states 0.5%. It also holds about 281 ETH deposited into Aave V3 for idle withdrawal ETH, a small dependency next to the TVL. An emergency pause path exists: a Pauser Safe calls
LRTConfig.pauseAll()and a Manager Safe can block user transfers. Kelp said it paused rsETH contracts across mainnet and several L2s during the April incident (source reliability: established publisher quoting the team). The pause was in place for weeks: a 2026-05-12 Kelp post only said withdrawals would be unpaused tentatively later (source reliability: unverified publisher). The runbook states that unpausing happens only on sign-off from the incident and security leads, and the cited docs give no maximum pause duration. Assessment: the Pauser Safe can halt exits with no stated bound, so exit availability depends on that sign-off. - The April 2026 shortfall. Kelp's official notice says assets backing rsETH minted on Ethereum Mainnet remained fully secure and that 116,500 rsETH was released from the bridge. The Aave risk provider's report says the adapter held 116,723 rsETH before the exploit and 223 after, and that remote-chain claims were 152,577 rsETH against 40,373 rsETH of confirmed backing (source reliability: community publication). That report estimated USD 123.7 million of bad debt on Aave if losses were shared evenly across all rsETH holders.
- How it was covered. A May report says 106,993 rsETH was recovered (89,567 from Aave and 17,426 from Compound liquidations) out of about 112,103 unbacked, with about 5,211 rsETH left for the DeFi United coalition (source reliability: community publication). A Kelp post of 2026-05-12 says 117,132 rsETH would be refilled into the OFT adapter over two weeks (source reliability: unverified publisher). A press report of 26 May says the final batch of 23,737.72 rsETH went to the adapter and rsETH is fully collateralized (source reliability: unverified publisher). Three further sources describe the refill as complete: a Blockworks summary of the Aave post-mortem says five tranches between May 13 and May 26 restored backing in full (source reliability: unverified publisher), and the Aave post-mortem on X and its May development update say the same (source reliability: unverified publisher; community publication). The post-mortem's tranche rows (25,000, 25,000, 20,000, 26,758.29 and 20,373.72 rsETH) add to 117,132.01 rsETH (my calculation), which is 1,000 more than its stated 116,131.72 total, and 117,132 matches the figure Kelp announced on 2026-05-12. These accounts all come from weak sources, so the completion of the refill is unconfirmed; the adapter balance has not been checked on-chain in the workspace.
- Exit on remote chains. Kelp's recovery guide (source reliability: official protocol blog) says the old rsETH on Base, Mantle and Ink was retired on 2026-10-05 when those networks moved to CCIP, and bridging of the old token on 22 networks (including Linea, Optimism and Zircuit) was retired on 2026-06-15; Arbitrum is the only network where the old rsETH still bridges. Holders of old tokens recover by burning them and paying 100 USDC per address per network, with settlement once a quarter and a deadline of 2027-06-15. One cohort has no standard exit: CCIP rsETH on Zircuit can no longer be bridged and is outside the recovery program, and holders are told to email the claims address. Assessment: this is a narrow, disclosed gap on one remote chain with no measured size, so it does not change the Ethereum exit picture, but it shows remote-chain exits depend on Kelp's manual process.
- Downstream bad debt today. The Aave Core pool shows a measured WETH deficit of 52,964 WETH at block 26379467. The rsETH reserve itself shows zero deficit and is frozen. These are Aave's books, not Kelp's. The file does not say how much of the WETH deficit comes from this incident, so it is not attributed.
- Loss handling documented by Kelp. No Kelp document in the workspace states a loss-socialization rule or a reserve for shortfalls. The July 2024 review notes no validator insurance. See Incentive Alignment for who paid.
Incentive Alignment
Public accounts of the April loss describe outside lenders and a coalition of ecosystem partners taking part in the recovery, with no Kelp-owned fund, so alignment between Kelp and its holders rests on goodwill and not on a stated rule; whether pledges were paid is unconfirmed.
- Who paid. Kelp's recovery accounting of April 24, as reported by an established crypto publisher, put the initial shortfall at 163,200 ETH, of which Kelp recovered 40,300 rsETH on its own. The Arbitrum Security Council froze another 30,700 ETH. Public commitments from Mantle, Stani, EtherFi, Lido and Golem totalled 43,500 ETH against the remainder, which was about 89,500 ETH at that date.
- Payout record. No payout row is recorded for the April 2026 incident. The established source shows recovered and frozen assets and partner pledges, and it states that the frozen ETH needed an Arbitrum governance vote for release. No non-weak source confirms that recovered funds or pledges were paid to users.
- Commitments are not payment. The April 24 figures are pledges. A later community report says the Aave positions were liquidated on 2026-05-06 and that the remaining rsETH gap would be covered by the coalition. (source reliability: community publication) No non-weak source confirms that every pledge was paid.
- No Kelp fund. No insurance, reserve, safety-module or compensation fund owned by Kelp appears in the workspace. Aave's Umbrella safety module belongs to Aave and backs Aave suppliers, not rsETH holders, so it is not recorded as a Kelp fund; the same article notes it does not cover L2 deployments. The
fundslist in the sidecar is empty for that reason; an empty list means none found, not proven absent. - Fee take versus risk. The protocol earns about USD 2.0 million of revenue per year on USD 34.6 million of fees, which is thin next to a USD 293 million loss. Assessment: the fee stream is small next to the loss, so a Kelp-funded reserve of that size would take many years of revenue at the current rate.
- Who can change the rules. The 6-of-11 Admin Safe holds the default admin role, and a 3-of-6 Manager Safe and a 1-of-5 pause-only Safe hold the other roles (source reliability: unverified publisher). The on-chain default admin is the 0xb3696a... address. The detailed control analysis belongs in the governance report.
- Past record flag. A Kelp post of 2026-09-15 mentions a 24-hour pause on one address (source reliability: unverified publisher); the underlying reason is not in the workspace, so it is noted without a conclusion.
Economic Attack Vectors
The vector that actually cost money was a bridge-verification failure on the Unichain-to-Ethereum route; the measured Arbitrum route is now 4-of-4 and the documentation states all routes are, but the exploited route's current configuration is unverified, and the structure that made it expensive (rsETH as collateral in lending markets, a shared lockbox behind remote copies) is still in place.
- Forged cross-chain message (realised, April 18, 2026). DefiLlama lists a Cross-Chain Message Spoofing hack of USD 293 million on Ethereum and Arbitrum. The loss was 18.41% of the USD 1.59 billion TVL of the day before. The Unichain-to-Ethereum route was configured as a 1-of-1 DVN path, and the forged packet had no source-side burn (source reliability: community publication). LayerZero's account says an attacker social-engineered a developer, poisoned the RPC nodes the DVN used, and that the LayerZero Labs DVN now refuses to sign as the sole required attestor (source reliability: community publication).
- Current bridge configuration. The docs state every rsETH LayerZero route now enforces 4-of-4 independent DVNs with 64 confirmations. An on-chain read at Ethereum block 26393314 (2026-11-12) matches for the rsETH adapter route to Arbitrum: verifier quorum 4 of a set of 4, 64 confirmations, and a configuration owner threshold of 3. The same record shows a configuration change delay of 0.0 hours and no rate limit on that route. The read covers only the Arbitrum route. The exploited Unichain-to-Ethereum route has no rsETH read in the route data, so its current configuration is not verified; the 4-of-4 statement for all routes rests on the documentation. The zero change delay on the Arbitrum route means the owner could loosen it without notice. Kelp's X posts say the migration to Chainlink CCIP finished on 5 October (source reliability: unverified publisher). The CCIP lock-release pool reads token buckets: the route to Mantle has 300 rsETH capacity refilling at 3,472,222,222,222,222 wei per second, about 300 rsETH per day (my calculation), while the Optimism route inbound bucket reads 42 rsETH with a refill of 1 wei per second. Some CCIP off-ramps are flagged as unread in that file, so the CCIP limits are partial. The dependency graph lists LayerZero v2 at 5.03% of TVL and Chainlink CCIP at 3.57%, both unresearched by us.
- Lockbox backing remote copies. An Ethereum lockbox holds the rsETH that backs L2 copies, and a forged message can drain it with no burn. After the CCIP move, a second lockbox exists for CCIP. The lockbox balance relative to remote supply is not measured here.
- Collateral use in lending markets. The attacker deposited 89,567 rsETH into Aave and borrowed WETH. The look-through is that any market that accepts remote rsETH inherits bridge risk. Aave's reserve cap for rsETH is 530,000 against 288,885 supplied. Concentration of 84% of Ethereum supply in one market makes a freeze there a liquidity event for most holders.
- Oracle manipulation. No cited route manipulates the core rate beyond the 1% bounded update, the stale-rate round trip in the audit finding. See Oracle Risk.
- Dependency failures. Slashing at EigenLayer and operator failure are not recorded as loss sources in the cited material, but the only cited statement on validator loss cover is that Kelp has no insurance for its validators (source reliability: established, July 2024). Whether EigenLayer slashing is live today is not verified in the workspace.
- Withdrawal run. A mass exit would use the 15-21 day queue or the 0.5% instant route. The workspace does not give the instant-route capacity beyond the "availability of funds" caveat, so a run cannot be sized.
- Prior family incident. DefiLlama lists a 2024-02-21 Gain infinite-mint of USD 3.01 million as a family match for another product. The target, scope and user loss are not verified, so it does not enter the rows in the sidecar.
Aggregator Scores
Two third-party scores exist and both are curated snapshots, and neither says anything specific about the April 2026 exploit.
- CertiK Skynet lists 78.8 with grade BBB, captured 2026-10-24 as a curated snapshot that is not guaranteed to reflect current state.
- Exponential lists 7.0 (Good), captured on 2026-04-13 from the Wayback Machine. That capture is five days before the April 18 exploit and its note says it is not guaranteed to reflect current state. It records no documented hacks and no death-spiral concerns, and says the multisig has fewer than four signers with no timelock information. Its hack statement is stale, so it is given no weight.
- Stader, the issuer of ETHx (11.06% of TVL), is a scored dependency (yellow, scored 2026-10-31, usable until 2027-04-29). Its report states ETHx has only about USD 100,000 of DEX depth within the 2% price-impact limit. (dependency research, background context only)
Data Completeness
The on-chain control reads, DEX depth, holder concentration, DefiLlama and docs are complete and fresh to within a few days, but verifier configuration beyond one LayerZero route, current lockbox balances, loss-fund status and the final recovery are not independently verified.
- Fresh and measured. DefiLlama TVL (2026-11-12), DEX depth (2026-11-12), holder concentration (block 26395177, 2026-11-12) and the asset composition (2026-11-12) are current. The on-chain control read is from block 26379467 on 2026-11-09 and its status is partial.
- Not measured. The current CCIP verifier set and the off-ramps flagged as unread, the lockbox balance against remote supply, whether EigenLayer slashing is live, the Aave price source for rsETH, and a per-day release limit on the LayerZero adapter route (none is recorded).
- Weak-source dependence. The recovery outcome and the CCIP migration rest on Kelp's and Aave's X posts, a Blockworks summary and a press article (source reliability: unverified publisher) and on risk-provider and Aave forum reports (source reliability: community publication). The refill tranches in the Aave post-mortem sum to 1,000 rsETH more than its stated total. Those are read as team claims.
- Dependencies. CCIP and LayerZero v2 are unresearched by us. Only Stader is scored. The chain profile lists Ethereum only, with no L2BEAT stage.
- Not applicable or absent. No Kelp-owned fund was found, so the
fundslist is empty. There is no fund-balance check file in the workspace, and no vote-concentration venues are recorded. - Audit scope. The audits in the workspace address contract code, not the off-chain verifier that failed in April. Audit findings on rate freshness are covered under Oracle Risk; other economic findings in the audit set are not reviewed here, so this is a gap.
Score: 4.9/10
Rubric security-score-v2.6; data completeness 1.00
- Solvency redemption
- 3.50 points
- Oracle design
- 7.50 points
- Dependency risk
- 5.50 points
- Liquidity concentration
- 1.00 points
- Yield design
- 5.50 points
| Factor | Level | Evidence | Weight | Points |
|---|---|---|---|---|
| Solvency redemption | L1 | supported | 0.35 | 3.50 |
| Oracle design | L3 | supported | 0.25 | 7.50 |
| Dependency risk | L2 | supported | 0.15 | 5.50 |
| Liquidity concentration | L0 | supported | 0.10 | 1.00 |
| Yield design | L2 | supported | 0.15 | 5.50 |
143 citations · click markers to view evidence
Team & Transparency
Social & Team Transparency: Kelp
Kelp publicly names two co-founders and a company, and it publishes contract addresses, audit links and incident write-ups. Its track record carries a $293M bridge exploit on April 18, 2026, plus a 2024 website attack and a family-product record from February 2024. How long withdrawals stayed halted after the 2026 exploit, and whether they fully recovered, rests only on unverified-publisher sources. Public code activity is thin: commits in 2 of the last 12 months. The evidence does not show completed compensation for every affected user, and it says nothing about community size or sentiment.
Team Identity
Two co-founders are named in official Kelp material, and a company is named, but its jurisdiction and the founders' prior work are not independently verified.
- Named founders. An official Kelp funding announcement names Dheeraj Borra and Amitej Gajjala as co-founders.
- Roles. The family's EU-style crypto-asset white paper lists Borra as Co-Founder and Chief Technical Officer. It lists Gajjala as Co-Founder and Chief Executive Officer. The white paper is a family-level (Kernel/Kelp) document, and current roles are not reconfirmed.
- Other named people. The same white paper lists Olivier Galaud as Chief Business Officer and Tim Wong as Business Development Lead.
- Prior projects. A Kelp blog says the founders "formerly" founded Stader Labs. This is a team statement with no independent source in the workspace. The same blog says Kelp and Stader are separate entities, so Stader's record is not counted here.
- Legal entity. Kelp's terms of use name Evercrest Technologies Inc. as the company that makes the app available. The terms state no jurisdiction of incorporation. The white paper gives a registered address in Panama City and names Stakewave Foundation as parent. A registered address does not by itself prove the place of incorporation.
- DAO structure. "DAO" is part of the brand. The sources do not show a legal DAO structure.
- Anonymity. The team is not anonymous.
- Investors. The $9M private-sale announcement names angel investors including Scott Moore (Co-founder, Gitcoin) and Sam Kazemian (Co-founder, Frax Finance).
Track Record
Kelp has operated since late 2023 but has three recorded incidents, one of them a large bridge exploit in April 2026. Operating age is code-derived from the family listing date of 2023-12-21 and the contract creation dates.
Core contracts (live Ethereum version). The official address list gives these fund-holding or fund-moving proxies. The deposit pool proxy was created on 2023-12-10 per the contract-creation record. Proxy creation does not date the implementation behind it.
Roadmap delivery. The sources do not give promised-versus-delivered milestones. Roadmap adherence cannot be assessed. Insufficient data for this assessment.
Incidents
April 18, 2026: rsETH bridge exploit (this deployment). Severity: major.
- Kelp's April 19 response says a forged packet released 116,500 rsETH from the Ethereum bridge lockbox. It attributes the compromise to third-party-operated verification infrastructure.
- DefiLlama records $293M lost, and its history gives previous-day TVL of $1,591,612,989 and a loss share of 18.41%. The loss share is above the 1% major threshold and below the 25% catastrophic threshold, so the incident is major. Whether any user loss stayed uncovered is unknown, so the separate catastrophic criterion of at least USD 50M uncovered user loss is neither shown nor ruled out; the TVL share alone does not settle it.
- Attribution is disputed. A LayerZero incident report (source reliability: community publication) blames Kelp's single-verifier setup. A press report (source reliability: unverified publisher) says Kelp claims LayerZero personnel approved that setup. Kelp's September 25, 2026 X post (source reliability: unverified publisher) says it filed a lawsuit against LayerZero and its co-founder. The X post is team content, and the lawsuit claim is not independently verified.
- Recovery: a Blockworks Intel summary (source reliability: unverified publisher) says Aave Labs-led coordination restored rsETH backing in five tranches totalling 116,131.72 rsETH between May 13 and May 26. No non-weak source confirms this, so completion is unconfirmed. Even if accurate, restored backing does not show how much of the cost fell on users, Kelp or partners. Completed user-loss coverage is unknown, so major is also the conservative rule-based default.
- Remediation: a Kelp X post (source reliability: unverified publisher) states a hardening pass (4 independent attestors, 64 confirmations). Official docs later state 4-of-4 independent verifiers on every route. The fix is team-stated; the current configuration of the exploited Unichain-to-Ethereum route is not verified in the workspace, so remediation of that route is probable, not verified. The statement that every route is covered is team-stated and not verified for all routes.
- Kelp's own account in the workspace is only the initial response, which says a root-cause analysis was requested from the provider. The workspace holds no later Kelp root-cause report, but that does not show none was published. A post-mortem by Aave Labs (source reliability: unverified publisher) exists and covers Aave's exposure and recovery; it is not a Kelp publication.
July 22, 2024: website (DNS) attack (this deployment). Severity: major as a rule-based default.
- Kelp's post-mortem says the dApp showed malicious wallet transactions because attackers took over the domain registrar account. It acknowledges user fund losses.
- Unknown facts: the loss amount, the incident-date TVL for affected users, and whether affected users were reimbursed. Major is the conservative rule-based default; it does not assert a proven loss threshold.
- The team says the issue was resolved within five hours and that it would leave the registrar. That is a team statement.
February 21, 2024: Gain infinite mint (family product). Severity: major.
- DefiLlama records a $3.01M infinite-mint on Gain, labelled a family product.
- The DefiLlama amount exceeds USD 1M, which meets the major criterion. Gain-specific TVL and reimbursement are unknown, and remediation is unknown. Kelp's TVL is not a Gain denominator, so no loss ratio is asserted.
- A targeted web search found no post-mortem, fix announcement or compensation statement for this incident. It also found that the DeFiHackLabs 2024 index (source reliability: unverified publisher) lists a GAIN exploit on 2024-02-21 with a loss of about 6.4 ETH. That figure conflicts with DefiLlama's $3.01M, and the search could not settle which is right. It also could not confirm from an authoritative source how the GAIN token in that index relates to Kelp's Gain product. DefiLlama's family match is the only link in the non-weak sources, and it is kept as the basis for the row.
DeFiHackLabs holds no matching records for this protocol.
Own-token depeg history
rsETH (Ethereum 0xA1290d69c65A6Fe4DF752f95823fae25cB99e5A7) is Kelp's issued token. The depeg measurement is unusable because no exchange rate was available. An empty run list does not show price stability. No 1:1 redemption statement was found, so rsETH is not marked as a fixed-par asset.
Stress and pauses
The sources do not allow a firm statement of how long exits were halted after the April 2026 exploit or whether they fully recovered, because every source on this point is an unverified publisher.
Kelp's May 12 X post (source reliability: unverified publisher) says withdrawals would be unpaused, tentatively within 24 hours, after the first tranche to the LayerZero OFT adapter. An Aave timeline (source reliability: unverified publisher) dates the reopening of Kelp withdrawals, bridging and claims to May 15. Assessment: these posts imply that withdrawals were paused from April 18 until mid-May, but a prospective unpause statement and a third-party timeline do not confirm the dates. Resilience is therefore left unscored for lack of a qualifying source.
Social Presence
Public development is sparse and official posting continues, while community size and sentiment are not measured.
- GitHub. Across five listed repositories, commits appear in 2 of the last 12 months, with the latest on 2026-02-10. The main
LRT-rsETHrepository shows one commit (February 2026); the December 2025 commits belong to the sibling Kred repository. Contributor count is not available. - Twitter/X. Kelp's X account posted on 2026-09-15 and 2026-09-25. The September 25 capture (source reliability: unverified publisher) shows 299 likes, 20 reposts and 29 replies. Follower counts are not available, and two posts are too few to judge engagement quality.
- Discord/Telegram. The links appear on the official blog footer. Membership and discussion quality were not measured.
- Blog. The official blog has dated posts including the Q1 2026 report, incident posts and the October 5, 2026 recovery guide. The sampled dates do not establish a monthly cadence.
Transparency Indicators
Kelp publishes mechanisms, addresses, audit links and a bounty, but its incident disclosure and financial reporting are incomplete.
- Addresses and code. The docs link the public codebase and list addresses per chain. The repositories are public, but the sources do not show that every running component is reproducible.
- Audits. The official audit page lists Sigma Prime, Code4rena, MixBytes and Bailsec reports. Audit scope and deployed-version coverage are assessed in the code-security report.
- Bug bounty. Docs describe a live Immunefi program of up to $250,000, with a $100,000 minimum critical payout. The Immunefi page shows the same maximum.
- Financials. Kelp's Q1 2026 report gives approximately $1M protocol revenue. No periodic treasury or expense reporting was found. DefiLlama shows trailing-365-day revenue of $1,983,970 and fees of $34,593,465.
- Incident reporting. The 2024 post-mortem is detailed. For April 2026, the first-party account in the workspace is the initial response only, and a third-party Aave Labs post-mortem (source reliability: unverified publisher) also exists. Whether Kelp later published its own root-cause report is not established. The Gain record has no identified post-mortem after a targeted search. Post-mortem coverage across incidents is therefore only partly established.
- Open source. Public repositories exist; the main code repository is public. Closed components are not identified in the sources.
Red Flags
- A $293M bridge exploit (18.41% of previous-day TVL) on the current deployment, with attribution disputed. Kelp's own X post (source reliability: unverified publisher) says it has sued LayerZero; that claim is unconfirmed. Final user-loss coverage is not documented.
- A prior website attack with acknowledged user losses and an unknown loss amount.
- A $3.01M Gain exploit recorded by DefiLlama for a family product, with no post-mortem found after a targeted search.
- Unverified-publisher sources report that withdrawals were paused for weeks after the April 2026 exploit; the duration is unconfirmed.
- Public commits are in only 2 of the last 12 months, and the last was in February 2026.
- Founder histories and the company's jurisdiction are not independently verified.
No sources show deleted content, fake engagement or unattributed copied code. Missing evidence is not a positive finding.
Positive Signals
- Two co-founders and further executives are named in official documents.
- Kelp responded publicly to the 2024 and 2026 incidents, published a 2024 post-mortem and described its 2026 hardening. No Kelp response to the Gain record was found.
- An unverified-publisher Blockworks Intel summary reports that rsETH backing was restored in full. This is unconfirmed and carries no weight here.
- Addresses, audit links and an active $250,000 bounty are published.
- Official posts continued through September and October 2026.
Data Completeness
Technical and incident evidence is stronger than team biographies, community metrics and compensation accounting. This analysis uses collected documents and captured pages, not live social media scraping.
- No article collection is present, so no Blockworks (ex-Messari) Protocol Overview or Quarterly Report applies. The Blockworks Intel event page cited above is a news summary (source reliability: unverified publisher), not a quarterly report.
- Missing: follower counts, Discord/Telegram size, contributor counts, independent verification of founder histories, the incorporation jurisdiction, a Gain post-mortem (a targeted web search found none), completed user compensation, and an rsETH exchange-rate series for depeg measurement.
- Inactive chains: none listed in the workspace for this report.
Score: 5.7/10
Rubric security-score-v2.6; data completeness 0.65
- Team identity
- 7.50 points
- Disclosure
- 3.50 points
- Development activity
- 5.50 points
| Factor | Level | Evidence | Weight | Points |
|---|---|---|---|---|
| Team identity | L3 | supported | 0.45 | 7.50 |
| Disclosure | — | not disclosed | 0.35 | 3.50 |
| Development activity | L2 | supported | 0.20 | 5.50 |
Track record: 4.9/10
Rubric security-score-v2.6; data completeness 0.80
- Operating history
- 7.50 points
- Incident record
- 3.82 points
- Resilience
- 3.50 points
| Factor | Level | Evidence | Weight | Points |
|---|---|---|---|---|
| Operating history (computed) | L3 | supported | 0.30 | 7.50 |
| Incident record (computed) | L1 | supported | 0.50 | 3.82 |
| Resilience | — | not disclosed | 0.20 | 3.50 |
47 citations · click markers to view evidence
Governance
Governance Analysis: Kelp
The rsETH contracts are controlled through Gnosis Safe multisigs; the sources read show no vote that binds them. Contract upgrades on Ethereum and asset onboarding wait 10 days in a timelock, and a 6-of-11 Safe proposes them. Other fund-affecting powers act instantly through 3-of-6 and 6-of-11 Safes, including fees, mint cap, bridge settings and the CCIP lockbox. Signer identities are not published, so the claim of external signers cannot be checked. This assessment rests on on-chain reads at Ethereum block 26379467 (2026-11-09) plus the project's own documentation.
Governance Structure
The protocol is a multisig-run system with a timelock on upgrades, and no binding token vote was found. Scope: rsETH and the Kelp restaking contracts on Ethereum, plus its L2 deployments. The Kernel and Gain products are not assessed here.
- Type: multisig governance. The documentation states "All authorities are Gnosis Safe multisigs."
- Voting mechanism: no vote that binds the rsETH control paths was found. A governance explainer for KernelDAO (a separate product) says "Vote (Coming Soon) Use your voting power to shape Kernel DAO" . The same article says $KERNEL holders can create and vote on proposals . It is an educational piece about KernelDAO, and no governor contract or Snapshot space appears in any measured rsETH control path. Whether any vote influences Safe actions on rsETH is not established by the sources read.
- Vote concentration: the measured vote file lists no venues (
venues: []), so there are no proposals, top-10 shares or quorum figures to report . - Recent proposals: none found.
- Timelock: the Ethereum ProxyAdmin is owned by a
TimelockControllerwith a 10-day delay . L2 timelocks are 3 days. On-chain, the Ethereum timelockmin_delay_secondsis 864000 (10 days) at block 26379467 . - Emergency bypass: a 1-of-5 Pauser Safe can pause; unpausing needs the admin Safe .
Admin Key Analysis
Admin power is held by three Safes with different thresholds, and no EOA holds upgrade rights. A 1-of-5 pauser, a 3-of-6 manager and a 6-of-11 admin Safe each act without a timelock for the functions listed below.
| Safe | Address | Threshold (docs) | Role per docs |
|---|---|---|---|
| External Admin | 0xb3696a817D01C8623E66D156B6798291fa10a46d | 6-of-11 | Core DEFAULT_ADMIN_ROLE, recovery custody |
| Admin | 0xb9577E83a6d9A6DE35047aa066E3758221FE0DA2 | 3-of-6 | Owns an Admin-Safe ProxyAdmin |
| Manager | 0xCbcdd778AA25476F203814214dD3E9b9c46829A1 | 3-of-6 | Day-to-day operations |
| Pauser | 0xb6AbB489aCA4583833230F10B3A7670114D09559 | 1-of-5 | Pause only |
The docs list these thresholds and roles: . The docs also state the upgrade path: . The on-chain read at block 26379467 confirms 6 of 11 owners for the External Admin Safe and 1 of 5 for the Pauser . The Manager Safe has threshold 3 with 6 owners and the Admin Safe has threshold 3 with 6 owners; none of the four has a Safe module or guard in the read. Signer addresses are public on-chain; names and affiliations are not. The docs claim external signers, but the available sources do not identify them, so that claim is unconfirmed.
What each key can do (read at block 26379467, 2026-11-09):
- External Admin Safe (6-of-11), no delay on these:
- Calls
setContract(bytes32,address),setTokenandsetRSETHonLRTConfig, which repoint the core contract registry . - Calls
recoverFrozenFunds(address)on rsETH . - Unpauses the deposit pool, oracle, withdrawal manager and rsETH, and sets
setPricePercentageLimitandupdatePriceOracleForon the oracle. - Holds
DEFAULT_ADMIN_ROLEonLRTConfigdirectly . Whether this role can grant other roles, and with what delay, is not established: the read lists thegrantRoleguard as unresolved. - Owns the CCIP rsETH lockbox pool and can call
transferLiquidity(address,uint256)on it at zero delay (read at Ethereum block 26379467) . ThewithdrawLiquiditypath is guarded by a rebalancer address that the read does not resolve. - Is also the proposer and canceller of the 10-day timelock.
- Calls
- Manager Safe (3-of-6), no delay:
- Sets
setMaxMintAmountPerDayfor rsETH,setProtocolFeeBpsand the instant-withdrawal fee, fee recipient and withdrawal delay. - Calls
blockUserTransfers(address[])on rsETH. - Holds the operator role, with two EOAs that also hold it:
0x7aad74b7f0d60d5867b59dbd377a71783425af47and0xf2099c4783921f44ac988b67e743daefd4a00efd. Operators callcompleteWithdrawalForUserandclaimStEth. - Is owner and delegate of the Ethereum LayerZero rsETH adapter (see Parameter Controls).
- Sets
- Pauser Safe (1-of-5): pauses via
pauseAllonLRTConfig. Its change history shows owner swaps on 2026-07-09 and 2026-07-23. - Timelock (10 days): proposer and canceller is the 6-of-11 Safe; executor is the zero address, meaning anyone may execute once the delay passes .
Benefit: a 1-of-5 pause lets the team respond to an incident within minutes. Cost: threshold-3 Safes and EOAs hold moving-funds or bridge-config powers with no delay.
- (actor)
- (actor)
- (actor)
- (governance)
- (contract)
- (contract)
- control
- control
- control
- control
Upgradability
Core contracts are upgradeable proxies whose upgrade path goes through the 10-day timelock. The deposit pool, oracle, converter, withdrawal manager, LRTConfig and rsETH each show the same ProxyAdmin 0xb61e0e39b6d4030c36a176f576aabe44bf59dc78, whose owner is the timelock 0x49bd9989e31ad35b0a62c20be86335196a3135b1 .
- Pattern: EIP-1967 proxy with ProxyAdmin (transparent style). Nothing was read as immutable.
- Fastest upgrade delay: 864000 seconds (10 days) for each of those six proxies, per the privileges table.
- Timelock history (block 26379467): 144 executed operations, none pending, and three delay changes. The delay was set to 180 seconds on 2024-05-06, 172800 seconds (2 days) on 2024-05-09 and 864000 seconds (10 days) on 2024-05-14.
- Latest upgrades: on 2026-01-24 a single timelock transaction upgraded the deposit pool, oracle, converter, withdrawal manager,
LRTConfigand rsETH. Earlier batches ran on 2025-12-20, 2025-09-26 and 2025-04-30. - Upgrades outside the timelock: upgrades dated before 2024-06-19 (for example the rsETH implementation changes on 2023-12-10 and 2024-02-14, and deposit pool upgrades in April 2024) carry no timelock in the record. That predates the 10-day delay and does not by itself show a bypass.
- Can upgrades change invariants? Yes. An upgrade can replace rsETH mint and balance logic. The 10-day window is the only user warning.
- Bridge contracts: the docs say some bridge roles, including the Arbitrum LayerZero owner/delegate and the CCIP owner/admin, are not under a timelock and are being migrated .
- Not assessed: the L2 chains were read on separate chains and are not traced here.
Parameter Controls
Most parameters change instantly by Safe signature, with limited on-chain bounds.
- Mint cap:
setMaxMintAmountPerDayis callable by the Manager Safe at zero delay. The rsETH view showsmaxMintAmountPerDayof 5000000000000000000000 raw units (5,000 rsETH if 18 decimals) . - Oracle rate step limit: the docs say
pricePercentageLimitis currently 1% . The 6-of-11 Safe can change this limit. - Fees:
setProtocolFeeBpsandsetInstantWithdrawalFeeare set by the Manager Safe at zero delay . At Ethereum block 26379467 the raw values areprotocolFeeInBPS500 onLRTConfigandinstantWithdrawalFee50 onLRTWithdrawalManager. The docs state a 0.5% instant-withdrawal fee , which matches the raw 50 only if the denominator is 10,000. The repository source (commit3dded885f6f7) capssetProtocolFeeBpsat a raw 1500 and the instant-withdrawal fee at a raw 1000, commented "Max 10%" . The sources read do not bind that repository version to the deployed implementations, so the caps and the percentage conversion of the protocol fee are unverified for the deployed contracts. - Asset onboarding:
addNewSupportedAsset(address,uint256)is restricted to the timelock role, with a fastest delay of 864000 seconds (10 days) at block 26379467 . - Bridge settings: the Ethereum LayerZero rsETH adapter
0x85d456b2dff1fd8245387c0bfb64dfb700e98ef3has owner and delegate equal to the 3-of-6 Manager Safe, config owner threshold 3, andconfig_change_delay_hours: 0.0at block 26393314 . This Safe can therefore change the verifier set of its own route without delay. The same row shows a verifier quorum of 4 from a set of 4, pinned to this app and not following the shared default, so a change to the LayerZero default set does not alter this route. The CCIP lockbox pool on Ethereum (0x55e5a21b4ccc7fa502434ab1109d4ede0397ab25) has the 6-of-11 Safe as owner and the 3-of-6 Admin Safe as limit admin at block 26393314 ; the owner can also move lockbox liquidity at once (see Admin Key Analysis). - L2 CCIP pools: the measured own-token burn-and-mint pools each have an owner with config owner threshold 3 and config change delay 0.0 hours. Owners: Optimism
0x0d30a563e38fe2926b37783a046004a7869ade6c(block 159410488) ; Mantle0xfa3ba682f2210d05d087b331c71f5b34f16cd3e9(block 103144737) ; Linea0xee68df9f661da6ed968ea4cbf7ec68fcfe375bc6(block 32577573) ; Base0x7da95539762dd11005889f6b72a6674a4888b56d(block 56148622) . The read does not show who the signers are, and the L2 timelocks are not traced here. - L2 LayerZero routes: the measured own-token LayerZero routes on Arbitrum and BNB Chain each list the same address as owner and delegate, with config owner threshold 3, config change delay 0.0 hours and a verifier quorum of 4 from a set of 4 that does not follow the shared default. Arbitrum: owner and delegate
0x96d97d66d4290c9182a09470a5775ff90daf922cfor two routes (block 523347312) . The docs list that address as the Arbitrum admin multisig, 3-of-6 . BNB Chain: owner and delegate0xb4222155cdb309ecee1ba64d56c8bab0475a95b0(block 133012081) . The read does not name the BNB Chain signers. - Harm potential: the oracle step limit and daily mint cap bound some changes. No on-chain bound on
setContractorrecoverFrozenFundswas read.
Decentralization Assessment
The protocol is centralized in operation, with a timelock as the main safeguard. Control sits in a few Safes. Upgrades and asset onboarding have a measured 10-day delay, while the other listed paths act without delay.
- Spectrum: closer to team-controlled multisig than to community governance. No binding token vote, proposal process or on-chain governor was found in the sources read.
- Roadmap: the docs state a migration of bridge roles to timelocks. No broader roadmap toward voting was found.
- Key-person risk: the 6-of-11 Safe alone can repoint the config registry, recover frozen funds and propose upgrades. The 3-of-6 Manager Safe alone can change bridge verifiers and set mint caps. Collusion or compromise of 3 Manager signers or 6 admin signers is the relevant scenario.
- Past bridge incident: an official project post says the bridge verification infrastructure was compromised on April 18, 2026, and 116,500 rsETH left the Ethereum lockbox . A community-publication source (source reliability: community publication) says the route used a 1-of-1 verifier; this historical configuration is unconfirmed . The current Ethereum route reads as a 4-of-4 verifier setting, but config changes remain instant.
- Token distribution: no governance token was found to control the rsETH contracts. The holder file shows rsETH holders, not voting power.
Governance Risks
- Instant admin paths: a 3-of-6 Safe controls mint limits, withdrawal fees, user transfer blocking and bridge verifier configuration without delay. This is the main risk.
- Registry bypass: the 6-of-11 Safe holds
DEFAULT_ADMIN_ROLEonLRTConfigdirectly and callssetContractat zero delay. Assessment: it can repoint core contracts without the 10-day wait. Whether it can also grant roles, and with what delay, is not established. - Lockbox liquidity: the same Safe can call
transferLiquidityon the CCIP rsETH lockbox at zero delay. - Operator EOAs: two EOAs hold the operator role and call withdrawal-completion and unstaking functions with no delay.
- Signer churn: the three Safes show recurring owner swaps in 2026 (for example 2026-07-23, 2026-09-08 and 2026-09-11). The new signers' identities are not identified in the sources.
- Voting attack surface: not assessable. No governor or Snapshot venue is measured for rsETH, so voter apathy and flash-loan voting cannot be rated; a binding vote would change this.
- Centralization creep: the Manager Safe's powers include the LayerZero owner and delegate roles. The docs say the bridge roles are being moved under timelock.
Data Completeness
Available: on-chain Safe, timelock and proxy reads at Ethereum block 26379467 (2026-11-09); bridge-route reads at block 26393314; the project's access-control and timelock pages. Missing:
- Signer identities for every Safe, and whether the claimed external signers are independent.
- Whether role grants on
LRTConfighave a delay. - Deployed fee caps and units: only repository caps (raw 1500 and 1000) were read, and they are not bound to the deployed implementation.
- Whether any vote influences Safe actions on rsETH; the measured vote file lists no venue.
- L2 contract control beyond the CCIP and LayerZero route owners listed above, which was not traced in this analysis.
- Signer identities behind the L2 and BNB Chain LayerZero and CCIP owners; their addresses and thresholds are measured, but the signers are not named.
Score: 3.7/10
Rubric security-score-v2.6; data completeness 0.75
- Admin control
- 5.50 points
- Change delay
- 1.00 points
- Decision process
- 3.50 points
| Factor | Level | Evidence | Weight | Points |
|---|---|---|---|---|
| Admin control | L2 | supported | 0.45 | 5.50 |
| Change delay | L0 | supported | 0.30 | 1.00 |
| Decision process | — | not disclosed | 0.25 | 3.50 |
53 citations · click markers to view evidence
Cross-Analysis
Cross-Risk Analysis: Kelp
Kelp's staking logic on Ethereum is the stronger part of the system. The main danger is the layer around it: bridged rsETH copies backed by lockboxes on Ethereum, one address that holds most of the Ethereum supply (inferred, not confirmed, to be a lending market), and Safes that change fund-critical settings with no delay. One of these failed in April 2026 for about $293M, and the exploited route's current settings were not read. Most recovery evidence comes from weak publishers, so the recovery is reported, not confirmed. Kelp is a liquid restaking protocol, not a bridge, so the "Conditions of Safe Use" section is omitted.
Products in scope: Kelp issues one token, rsETH, on Ethereum and as bridged copies on other chains. The Gain and Kernel products of the same team are outside the DefiLlama record and are not assessed. DefiLlama TVL is $739.06M, all on Ethereum . The sections below use "estimated" or "unconfirmed" where the underlying evidence is weak.
Top Risks for a Holder
1. Bridged rsETH (all remote-chain copies, plus any lender that accepts them): a repeat forged message empties the lockbox behind the affected route
- What can go wrong. Remote copies of rsETH are backed only by rsETH locked on Ethereum. A forged or wrongly verified message releases it with nothing burned on the source side, and lenders that took the copy as collateral then hold unbacked tokens. This happened on April 18, 2026: Kelp's own post says . The attacker then put 89,567 rsETH into Aave and borrowed WETH .
- Why it persists (cross-domain). An April incident account says the verification layer sits outside every audit scope . The collected audits address contract code, not the off-chain verifier that failed in April, and coverage of that layer is unconfirmed rather than shown absent . The code report finds no confirmed audit scope for the pool, bridge and cross-chain rate contracts . The 3-of-6 Manager Safe owns the adapter and can change its verifier set with no delay .
- Magnitude. Realised loss was $293M, which is 18.41% of the previous day's TVL . Today the two Ethereum lockboxes hold 4.99% and 3.55% of Ethereum rsETH supply . That is about 8.5% of the 340,581 rsETH supply, roughly $63M at $2,180.87 per rsETH (my calculation). Aave's April bad-debt estimate was $123.7M to $230.1M, depending on loss sharing .
- Likelihood: medium. The measured Arbitrum and BNB Chain LayerZero routes now read 4-of-4 verifiers with 64 confirmations . The CCIP routes carry token buckets that cap daily release. Against that, the exploited Unichain route was not read, the CCIP verifier sets and several off-ramps are unread, and the Arbitrum LayerZero route's release limit is unmeasured (the field is null, which is unknown). Reading the exploited route's current settings, and the CCIP risk-management settings, would move this to low or high.
- Outcome of the April recovery is unconfirmed. Several weak publishers say the lockbox was refilled in full. The accounts disagree on the final tranche (20,373.72 rsETH in one, 23,737.72 in another) and the post-mortem's tranche rows exceed its stated total by 1,000 rsETH .
2. rsETH on Ethereum (about 84% of supply in one proxy, inferred to be a lending contract): exit freeze and a thin market
- What can go wrong. In a stress event, most holders cannot exit through the market, and the native exit is slow or can be paused without a stated limit. The top holder owns 84.09% of supply . The economic report infers this is the Aave V3 rsETH aToken , and the overview says the holder's identity is not established beyond its proxy label . If the inference is wrong, the 84% concentration still stands as one address, but its lending-market exposure does not. Aave's rsETH reserve is frozen at the latest read . A freeze disables new supply and borrowing while existing positions stay open to repayment and liquidation . The same read shows the reserve not paused, with nearly all supplied rsETH available . The freeze alone is therefore not shown to block withdrawals; only a separate pause or exhausted liquidity would.
- Cross-domain link. DEX depth is about $0.5M at 2% impact against $737M of supply, and a $2M trade moves price 26.83% . The 15-21 day standard queue is the only exit not tied to that market or to the "subject to availability" instant route . The Pauser Safe (1-of-5) can halt everything, and unpausing needs a 6-of-11 Safe with no stated maximum duration . TVL already fell $240.4M in 30 days .
- Magnitude. Loss of access, not loss of principal: if the inference holds, about 286,400 rsETH (about $625M, my calculation) sits in one market. Weak sources suggest withdrawals stayed paused from April 18 to about May 15 . That duration is unconfirmed. A holder forced to sell into the market in the stress case faces a discount that grows quickly above $0.5M.
- Likelihood: high that the market exit is impaired in stress (measured depth). Medium that a pause or freeze recurs (it happened in April). The identity of the 84% holder (inferred, not established) would change the sizing.
3. All rsETH on Ethereum: instant-effect Safe powers bypass the 10-day timelock
- What can go wrong. A compromised or colluding Safe changes fund-critical settings at once. The 6-of-11 Safe calls
setContractat zero delay, which repoints the core registry, and can move CCIP lockbox liquidity . The 3-of-6 Manager Safe sets mint caps and fees, blocks user transfers and owns the bridge adapter. Two single-key accounts hold the operator role . - Cross-domain link. The code report records an acknowledged audit finding on the same dependency: a compromised manager can stop minting by setting the daily cap to zero (an estimated Medium severity; the audit gave no rating) . Signers are unidentified, the claimed external signers are unconfirmed, and the Safes show owner changes in July and September 2026 . The 10-day timelock applies to upgrades and asset onboarding, not to the other listed paths .
- Magnitude. The CCIP lockbox is 3.55% of supply (about $26M, my calculation). A registry repoint could in principle reach the whole $739M, but no source measures an on-chain bound on
setContractorrecoverFrozenFunds. Assessment: the worst case is theoretical until that bound is read. - Likelihood: low. It needs 3 or 6 signers to be compromised or collude, and no such event is in the record. The bridge-settings delay is still zero, so this is a standing exposure and not a one-off.
Combined Attack Vectors
Vector 1: Verifier weakening or failure → lockbox drain → lending bad debt → exit freeze
- An attacker or a compromised Manager Safe lowers the verifier set on a route, or compromises a verifier's infrastructure as in April (LayerZero says the DVN's RPC was poisoned ).
- A forged message releases rsETH from the lockbox behind the affected route (LayerZero or CCIP) with no source-side burn.
- The attacker deposits the released rsETH in a lending market, borrows WETH, and utilization reaches 100%, trapping depositors .
- If Kelp pauses contracts and lenders freeze the reserve, as accounts of April describe, remote-chain copies stay undercollateralised until a refill.
- Domains: code (verification layer: audit coverage unconfirmed), governance (zero-delay bridge config), economic (collateral use, lockbox backing), social (disputed responsibility).
- Basis: the documented 4-of-4 setting lowers single-verifier risk, but the exploited route was not re-read and a 3-of-6 Safe can change it instantly.
- Likelihood: medium
- Impact: severe
Vector 2: Stale-rate round trip on instant withdrawal
- rsETH's rate lags because the oracle reverts when the daily fee-mint limit is reached, an acknowledged High audit finding (unverified audit publication provenance) .
- An audit describes a possible round trip: deposit at the stale low price, force an update, withdraw instantly at the new price. It pays only if a stale-price window exists, the rate step exceeds the 0.5% instant fee and instant-withdrawal liquidity is available .
- The code check finds the described code unchanged in the verified deployed source (a limited signal: it shows the code is still present, not a confirmed vulnerability, and no exploit was tested) for the deposit pool and oracle . The Manager Safe can change fees and the withdrawal delay at once, which can widen or close the window.
- Domains: code (acknowledged findings), economic (fee and oracle bounds), governance (instant fee control).
- Illustration only (my calculation): if the 5,000 rsETH daily mint cap (about $10.9M) applied, a 1% rate step and the 0.5% instant fee would leave about 0.5%, roughly $55k per day. This is not a ceiling. The Manager Safe can change the mint cap and fees at once , and the audit appendix notes that a manager-only update function can bypass some price-threshold restrictions . The cost falls on remaining holders through the rate.
- Basis: it needs a stale-rate window and a rate step above 0.5%; no occurrence is in the record.
- Likelihood: low
- Impact: minor
Vector 3: Run on an illiquid, concentrated token during a market drop
- About 84% of Ethereum rsETH sits in one proxy, which the economic report infers is the Aave lending contract (not established), and users loop rsETH against ETH at up to ~14x leverage .
- A price move, a bridge scare or a stETH or ETHx problem triggers liquidations and exits.
- Market depth is about $0.5M at 2% impact, so forced sales push the market price away from the 1.084 ETH rate. The standard queue takes 15-21 days, and the instant route depends on available funds.
- A pause by the 1-of-5 Pauser Safe stops native exits. A lender freeze alone does not stop withdrawals; a lender pause or exhausted liquidity would, and no source shows either at the latest read.
- Domains: economic (liquidity, leverage), governance (pause control), social (April pause record).
- Basis: depth is measured thin and concentration is inferred; the trigger is external. Impact is loss of access plus a market discount.
- Likelihood: medium
- Impact: severe
Vector 4: Instant registry repoint through the 6-of-11 Safe
- Six signers of the External Admin Safe are compromised or collude.
- The Safe calls
setContractorsetTokenonLRTConfigat zero delay, repointing the contracts the deposit pool and oracle read. - The 10-day timelock, which applies to upgrades and asset onboarding, does not apply, and no on-chain bound on these calls was read.
- Domains: governance (zero-delay path), code (acknowledged centralization findings from MixBytes and an unverified-publisher analysis ), social (signers unidentified).
- Basis: it needs 6 of 11 signers, and no event is recorded. Impact is catastrophic only if the path has no bound; the missing bound is the fact that would settle this.
- Likelihood: low
- Impact: catastrophic
Vector 5: Migration confusion plus a website takeover
- Old and CCIP rsETH need different exits, and retired copies need a paid recovery that settles quarterly until June 15, 2027 .
- In July 2024 attackers took over Kelp's domain registrar account and the dApp showed malicious wallet transactions .
- Assessment: a similar takeover during the recovery window would meet holders who are already unsure which route is real. No source shows current registrar controls or a new compromise.
- Domains: social (website incident record), economic (manual recovery and stranded cohorts), overview (route inventory conflict).
- Basis: needs a fresh website compromise; the 2024 event does not show current controls are weak.
- Likelihood: low
- Impact: moderate (individual wallet losses; the affected population is not measured)
Systemic Risk Factors
- Backing assets. DefiLlama's dependency data shows stETH at 49.77%, ETH at 39.15% and ETHx (issued by Stader) at 11.06% of backing . The graph marks stETH and ETH "unresolved". Route to the dependency-risk owner: stETH is half of backing and no published report for it is in the dependency index.
- Stader (ETHx). Stader has a usable report: score 5.7, yellow, researched 2026-10-31, usable until 2027-04-29 . Its published cross-analysis (background context only) names control concentration on ETHx as its most serious risk: a 3-of-6 Safe can pause ETHx and enable Safe Mode with no delay, and a single key also controls the rate path. It also reports about $100,000 of ETHx depth within the 2% impact limit and that the Safe can upgrade the
ETHxPriceOracleproxy with no delay. Stader's report treats a freeze or discount as likelier than a loss. Kelp's economic report adds that the Stader exchange-rate bound is loosenable by a 3-of-6 Safe with zero delay . TheEthXPriceOraclefeeds Kelp's rate, so an ETHx rate fault enters the rsETH rate for 11.07% of TVL. - Bridges that carry rsETH. LayerZero v2 is 5.03% and Chainlink CCIP is 3.57% of TVL, both unresearched . At $739.06M TVL that is about $37.2M and $26.4M (my calculation), which matches the roughly 8.5% of supply held in the two lockboxes (Top Risk 1). The dependency index shows no published report for either bridge, so route settings cannot be compared with a bridge's own conditions of safe use. The route data is marked completed, but some components were not read. Settings as read (a null is unknown, not met):
| Route (framework) | Block | Quorum / set; confirmations | Pinned, not default | Daily share | Owner threshold; change delay |
|---|---|---|---|---|---|
| Ethereum to Optimism (CCIP) | 26393314 | null; null | no (follows default) | 0.0 | 6; 0.0 h |
| Ethereum to Mantle, Ink, Linea, Base (CCIP) | 26393314 | null; null | no | 0.024795 | 6; 0.0 h |
| Mantle, Linea, Base to Ethereum (CCIP) | 103144737; 32577573; 56148622 | null; null | no | 0.024795 | 3; 0.0 h |
| Optimism to Ethereum (CCIP) | 159410488 | null; null | no | 0.0 | 3; 0.0 h |
| Ink to Ethereum (CCIP) | not read | unknown | unknown | unknown | unknown |
| Ethereum to Arbitrum and BNB Chain; Arbitrum and BNB Chain to Ethereum (LayerZero) | 26393314; 523347312; 133012081 | 4 of 4; 64 | yes | null (unknown) | 3; 0.0 h |
- Readings are per chain at the listed blocks, not one synchronized snapshot. A rounded daily share of 0.0 does not mean no burst capacity: the Optimism inbound bucket reads 42 rsETH refilling at 1 wei per second, and the Mantle bucket 300 rsETH refilling about 300 rsETH per day .
- CCIP verifier fields are null and three CCIP off-ramps are unread on most routes, so no CCIP route can be called verified. The Unichain-to-Ethereum LayerZero route that was exploited has no row in the route data.
- Remote rate delivery. The L2 rate still travels as a LayerZero V1 message, so moving token transfers to CCIP does not by itself move the price path; the Chainlink rate-feed replacement is described as in audit .
- Layers with no usable report. Lido stETH (49.77% of backing), EigenLayer and Aave have no usable report in the dependency index. Aave is both an exit-liquidity venue (about 281 ETH of idle withdrawal ETH at block 26379467) and the venue most exposed to rsETH collateral. Route all deposit-path layers to dependency-risk reassessment; this synthesis adds no score.
- Warnings in the dependency graph. The graph lists no warnings (no circular backing and no downstream holders) as of 2026-11-12 . The graph does not list Aave, although Aave probably holds most Ethereum supply (inference), so the downstream concentration is outside the graph.
- Chain-level. DefiLlama counts Ethereum only. Remote copies live on L2s whose sequencer and bridge assumptions were not assessed; Zircuit CCIP rsETH holders have no standard exit . A chain-shutdown gap is recorded for Blast.
- Macro. A stETH depeg or ETH price fall hits backing value and loop positions. About $139.8M of the 7-day TVL drop was price effect .
- Correlation. A bridge scare, a lender freeze and a pause hit together: the same lockboxes, the same lender (inferred) and the same Safes appear in vectors 1, 3 and 4.
Risk Amplifiers
- Zero-delay bridge config + verification layer with unconfirmed audit coverage + 3-of-6 threshold. The setting that decides bridge safety has no confirmed audit coverage in the collected audits and no delay.
- 10-day upgrade warning versus 15-21 day exit. A holder who sees a harmful upgrade queued cannot finish a standard withdrawal before it executes . The bridge and registry paths give no warning at all.
- Concentrated collateral + thin market + frozen reserve. A freeze stops new supply and borrowing, not existing withdrawals, so it narrows the lender's role as a buyer of new rsETH. If a pause or exhausted liquidity blocked an exit in one venue, no deep alternative exists.
- Leverage loops (about 14x) + lender freezes. Looping users cannot add new supply or borrowing on a frozen reserve and stay exposed to repayment and liquidation .
- Unidentified signers + instant paths + signer churn. The signers' independence cannot be checked, and owners changed in 2026.
- Acknowledged audit findings + unchanged deployed code. The described rate-freshness code is still present in the verified live source; this does not show the findings are exploitable.
- Thin revenue + no loss fund. DefiLlama revenue is $1,983,970 over 365 days against a $293M loss . Another large shortfall would again need outside support.
- No Kelp-owned loss fund + disputed responsibility. Losses fall on lenders and a partner coalition; Kelp's lawsuit claim against LayerZero is unverified.
Risk Mitigators
- Rate design. The rate is computed from held balances, not a market price. A 1% per-update bound and a downside auto-pause apply . The proof-of-reserves reader differed from the oracle by about 0.004% at the read.
- Upgrade delay. Core upgrades have a measured 10-day timelock, with a 6-of-11 Safe as proposer .
- Review depth. MixBytes, Sigma Prime and Bailsec reviewed the core contracts; no critical group is unresolved. The collected audits do not tie their scope to the bridge verification path, so coverage of that path is unconfirmed .
- Mint caps and pause. Daily mint caps (5,000 rsETH on Ethereum, 500 per L2 pool) limit the speed of abuse, though the Manager Safe can change them at once, and a pause path exists. A community publication reports that the emergency multisig froze core contracts 46 minutes after the drain (source reliability: community publication; not independently confirmed) . Treat the speed as a reported response, not a demonstrated mitigation.
- No own-token backing. Measured backing is stETH, ETH and ETHx, with no KERNEL or rsETH in it, so a KERNEL collapse does not directly impair rsETH backing .
- Documented hardening. Kelp states 4-of-4 verifiers on all LayerZero routes; the measured Arbitrum and BNB Chain routes match, and the claim for all routes is team-stated and not verified. The Immunefi bounty is up to $250,000, small beside a $293M loss.
Contradictions & Inconsistencies
- Verifier set. Kelp's docs say every LayerZero route is 4-of-4. Community accounts say the exploited route was 1-of-1, which Kelp's post does not state. On-chain reads cover only the Arbitrum and BNB Chain routes, so "all routes" is not measured.
- Route inventory. The docs list LayerZero for Base, Ink and Mantle, while Kelp's October 12, 2026 post says those moved to CCIP on October 5, 2026 . The route data lacks LayerZero rows for those chains.
- Pause duration. The overview cites a 46-minute freeze from a community publication (unconfirmed). The economic and social reports say exits likely stayed paused for weeks, based on unverified-publisher posts. Both can hold: a fast freeze and a long pause.
- Lender status. The overview treats Aave's freeze as the April 20 state and unconfirmed today. The economic report reads the Aave rsETH reserve as frozen on 2026-11-09 . The later measured read applies.
- Audit scope. The economic report (citing a source from April) says the verification layer is outside every audit; current coverage is unconfirmed in the collected audits. The code report lists a May 2026 Bailsec review of an off-chain configuration script. That script reviews configuration, not verifier infrastructure. The two are not directly comparable.
- Supply figures. The overview uses 345,124 rsETH (block 26379467); the economic report uses 340,581 (block 26395177). These are different blocks, and the drop between them was not explained in the sources.
- Final refill tranche. Two weak sources give 20,373.72 and 23,737.72 rsETH. Neither is confirmed on-chain.
- Proposer of the timelock. The docs say proposers are "respective admin multisigs"; the on-chain read shows the 6-of-11 Safe, not the 3-of-6 Admin Safe. Governance and overview agree on the on-chain reading.
Overall Risk Profile
Kelp's core Ethereum mechanism (balance-based rate, bounded updates, 10-day timelock on upgrades and asset onboarding) is sounder than the layer around it. The scores in the worker reports reflect this: code 5.0, economic 4.9, social 5.7 (track record 4.9) and governance 3.7, out of 10. The most likely failure mode is a liquidity and exit event: a thin market, an 84% concentration in one proxy (inferred to be a lender), a frozen reserve that blocks new supply but not shown to block withdrawals, and a pause with no stated limit. The worst case is a second bridge verification failure or a compromise of the Safes that set bridge verifiers, which drains the lockbox, creates lender bad debt and strands remote holders; the April loss of $293M shows the scale. Against peers, the workspace holds no peer data, so no ranking is supported. What can be said: Kelp has a realised bridge loss and instant admin paths that a strong timelock would remove, and its recovery evidence is mostly weak-source.
Data Gaps
-
Some audit reports were unavailable or could not be read in full.
-
Blast was not checked: chain shut down (announced 2026-10-02; source: https://x.com/blast/status/2106032805280891073).
-
Some source material could not be collected; the report may be incomplete.
-
Some protocol documentation could not be read because the site could not be reached.
-
Some source material is unavailable because it was not found publicly.
-
Current deployment addresses could not be confirmed for some audited contracts.
-
Source summaries cover only the first 80000 characters of 2 document(s); later content was not assessed.
-
The exploited Unichain-to-Ethereum route has no current read. The "4-of-4 on all routes" statement rests on documentation.
-
The LayerZero v2 and Chainlink CCIP reports do not exist in the dependency index (status unresearched), so no bridge-side conditions could be compared. The bridge route read is marked completed, but CCIP verifier settings are null, three CCIP off-ramps are unread on most routes, and the Ink route is unsupported by the reader.
-
Stader is the only material dependency with a usable report. Lido stETH, ETH, EigenLayer and Aave have no usable report in the index, and the LayerZero v2 and Chainlink CCIP entries are unresearched (marked non-material there, though they carry 5.03% and 3.57% of TVL).
-
Current slashing activation at EigenLayer, operator allocations, L2 rate-consumer freshness checks and current website-domain controls are not established.
-
The lockbox balance against remote supply is not measured, so the current shortfall or surplus of bridged rsETH is unknown. Weak sources report the April recovery; its outcome remains unconfirmed.
-
The identity of the 84.09% holder, Aave's rsETH price source and Aave's current state beyond 2026-11-09 are unmeasured.
-
Signer identities and independence for every Safe, and any on-chain bound on
setContractandrecoverFrozenFunds, are unread. -
No audit states a confirmed revision, so equivalence between audited and deployed code is not established. The code report's direct checks of findings against verified deployed source stand with their limited meaning: they show whether the described code is still present, not a confirmed vulnerability, and no exploit was tested .
-
No web research was run for this synthesis; all facts come from the accepted reports and bundled data.
72 citations · click markers to view evidence