StakeWise V3

Rejected venues wait the longest for re-review; a rejection has to earn another look before the scheduled date.

The first protocol promoted out of the universe registry for a full review, and the review kills its own headline. StakeWise V3 (Ethereum mainnet since 2023-11-28, $702.9M protocol TVL at the 2026-08-13 review against Rocket Pool’s $0.99B and Lido’s $17.9B) is the permissionless-vault liquid staking design: anyone can launch a vault, choose its node operators, fee, and MEV policy, keep it private, and mint osETH against the stake at a 90% loan-to-value cap, with permissionless redemption against any position past 91.5% and liquidation past 92%. The engineering record is genuinely strong: five years on mainnet across V2 and V3 with no protocol exploit, no known slashing, and no osETH liquidation ever recorded; audits by Halborn (2023, twice), Sigma Prime (2023 and 2024), and ABDK (v4.0); and a clean peg through the 2025-11-03 Balancer exploit, where the emergency multisig recovered 73.5% of the 6,851 osETH taken (about $20.7M) within 90 minutes and vault backing was never touched. Rejected anyway, on the criterion the Ethereum LST category review actually used: realized validator distribution. The pitch is operator choice; the reality is that the Genesis vault holds roughly 40 to 45% of all protocol stake, and at the last verified count (LlamaRisk, 2024-06) StakeWise Labs itself ran 45.85% of that vault’s validators. Rocket Pool spreads stake across roughly 2,000 independent operators. A protocol whose flagship differentiator is decentralized operator choice is, in realized numbers, more concentrated than the incumbent it would displace from our list. Three further findings independently block approval. The second-largest vault, 75,533 ETH at a 15% fee, is unidentified, and we do not approve a venue where ten percent of protocol stake answers to an operator we cannot name. The docs describe a DAO lever that can raise a vault’s osETH mint cap to 99.99% LTV, the same governance-can-raise-LTV shape that produced Aave’s $196M rsETH bad-debt event; whether any vault holds that tier today is unverified, which is itself the problem. And the osETH exchange rate depends on a permissioned oracle set of 11 entities whose V3 source code was private at the last public review, behind a bug bounty capped at $200k against roughly $700M at stake. The bench note is genuine. Per-vault risk isolation, solo-staker osETH minting, and private allowlisted vaults are capabilities neither Lido nor Rocket Pool offers, and Lido copying the design for its V3 stVaults is the market agreeing the capability matters. The announced Obol DVT migration of the Genesis vault (2025-04) attacks the concentration finding directly. If the numbers move, this file reopens on evidence.

Kill criteria: the events that revoke this verdict
• Reopen only if every condition below passes together; another Ethereum staking provider failing does not substitute for StakeWise evidence
• A current reproducible on-chain snapshot shows Genesis below 30% of protocol stake, StakeWise Labs below 20% of Genesis validators, and the Obol migration complete
• Every vault above 5% of protocol stake publishes its operator identity, fee, admin, upgrade path and realized validator distribution
• No live vault uses the 99.99% osETH mint tier unless the executed governance record, collateral buffer and liquidation test are independently approved
• The live V3 oracle implementation, source, signer set and exact voting threshold are published and mapped to the deployed contracts
• Every live Genesis, osETH, oracle and optional v4 contract maps to a read audit report, and a proposed-size direct and secondary exit executes below 50 basis points

The research file

The mechanism

StakeWise V3, per its docs, is a modular collection of smart contracts for liquid staking on Ethereum and Gnosis Chain, built around customizable vaults, including blocklist, private, and ERC20 variants. Vault creation is permissionless: Chorus One’s guide describes vaults as non-custodial staking mini pools anyone can launch, with the operator free to choose client software, KYC allowlists, MEV relays, and DVT middleware, and private vaults accepting deposits only from addresses the vault operator whitelists. A depositor’s claim is a share of one specific vault’s staked ETH, not of a pooled validator set; rewards and penalties remain isolated to the individual vault, per the same guide. That is the structural difference from Lido and Rocket Pool: vault choice is the risk decision. The launch announcement (StakeWise on Medium, November 2023) frames V3 as a marketplace of staking offers, each vault with its own operators, performance, MEV strategy, and commission, and solo stakers can run a private vault and mint osETH against their own validators while keeping custody. Vaults v4.0, an optional upgrade shipped 2025-10-01, adds 0x02 maxEB validators, faster withdrawals, consolidations, and MetaVaults that spread stake across up to 50 sub-vaults; StakeWise says all v4.0 contracts passed an ABDK Consulting audit, a claim this review did not confirm against the report itself.

osETH is minted by staking assets into a vault and issuing osTokens against the stake as collateral, capped at 90% loan-to-value for standard vaults. The docs also describe a DAO-approved tier at 99.99% LTV, gated on at least 10,000 ETH staked, fees at most 5%, median-or-better performance, and 5M SWISE locked as slashing insurance; whether any vault operates at that tier today is unverified. LlamaRisk’s assessment of 2024-06-25 describes the live system as a maximum 90% LTV enforced in each vault, which keeps a minimum 3.2 ETH buffer per validator to absorb staking penalties. The DAO sets LTV thresholds and can raise or lower a vault’s minting threshold. osETH is a repricing token: its redemption value rises as rewards accrue, computed by the osTokenVaultController contract, with oracles setting avgRewardPerSecond to track the highest-LTV vault’s APY. The ratio is defended by a redemption mechanism past 91.5% LTV, where anyone can burn osETH from the position to bring it back to 90%, and liquidation past 92% at a 1% premium; LlamaRisk confirms the numbers and adds that redemptions hit the worst vaults first.

Slashing splits cleanly. Per the Chorus One guide, osETH holders remain protected from slashing losses by the overcollateralization buffer, while stakers who mint osETH keep the slashing risk of the vault they staked in. The docs put it as a standing reserve of ETH that absorbs slashing losses before osETH holders are affected: slashing hits the vault’s depositors first, and osETH takes losses only if a slashed vault’s losses exceed the buffer, at which point redemptions and liquidations shrink the osETH minted from that vault.

Who controls it

The primary source for this section is LlamaRisk’s collateral risk assessment of 2024-06-25, a two-year-old snapshot; items may have changed. Governance runs through the SWISE token via Snapshot votes executed on-chain through SafeSnap: a 3M SWISE threshold per vote, a 200K SWISE SafeSnap bond (about $6,000 at the time), a 24-hour question timeout and 24-hour cooldown, and roughly four proposals per month since the V3 launch. The token split is 25% investors, 25% founders, 50% community, on 48-month vesting.

Core protocol contracts are owned by a 4-of-7 multisig (Ownable2Step); LlamaRisk named the signers as Ali Nuraldin, Andreas Dittrich, Auryn Macmillan, Dmitri Tsumak, Gleb Dudka, Mikko Ohtamaa, and Otto de Voogd, with the Genesis vault using a separate 1-of-2 multisig for reward splitting. Most core contracts are immutable, but individual vaults use the UUPS upgradeable pattern and each vault admin chooses whether to take upgrades, so a depositor’s contract risk includes their vault admin’s upgrade decisions; v4.0 is explicitly optional. There is no pause mechanism: LlamaRisk writes that the lack, while strengthening permissionlessness, severely limits any intervention, and osETH minting can be throttled only indirectly by adjusting capacity or removing vaults from the registry. No protocol timelock exists beyond SafeSnap’s 24-plus-24-hour governance delays.

The osETH exchange rate comes from a permissioned oracle network of 11 entities voting every 12 hours: Chorus One, StakeFish, Deutsche Telekom, Finoa, Bitfly, SenseiNode, Gateway.fm, Gnosis Chain, P2P, DSRV, and StakeWise Labs, per LlamaRisk, which states the voting threshold as 7 of 11 in one passage and 6 of 11 in another. The V3 oracle service source code was private at the report date; only V2’s was public.

The emergency capability is demonstrated, and an advisor should read it both ways. During the 2025-11-03 Balancer exploit the StakeWise DAO executed emergency recovery 90 minutes after the initial exploit using its multisig, per November 2025 reporting by coingape.com, coinedition.com, and crypto.news. That is fast, competent response, and it is also proof that a multisig can move user-adjacent funds without a vote when it decides the situation warrants it.

The record

The StakeWise mainnet pool launched in March 2021 with the V2 dual-token design, sETH2 for principal and rETH2 for rewards. V3 launched on Ethereum mainnet 2023-11-28, and Vaults v4.0 on 2025-10-01. DefiLlama (pulled 2026-08-13) shows $702.9M of TVL, $701.8M on Ethereum and $1.1M on Gnosis, against a peak of $1.58B on 2025-10-07; the trajectory ran $236M in January 2024, $482M in January 2025, $1.10B in January 2026, and $581M in July 2026. The DefiLlama series starts 2021-01-26, so the slug carries V2 and V3 history combined. Our own 2026-08-14 universe survey recorded about $369M across two tracked pools; that is pool-level yield-tracker coverage, not protocol TVL, and both numbers are reported here.

On losses: LlamaRisk wrote in June 2024 that to its knowledge no validator operated on behalf of StakeWise had been slashed, and per IntoTheBlock RiskRadar data no osETH position had ever been liquidated. No exploit of StakeWise contracts was found in any source reviewed. The material incident is external: on 2025-11-03 Balancer V2 composable stable pools were exploited for about $128M across chains, taking roughly 6,851 osETH plus 13,495 osGNO from Balancer pools. StakeWise’s emergency multisig recovered about 5,041 osETH (about $19M) and the osGNO (about $1.7M), 73.5% of the stolen osETH, within about 90 minutes; the remaining 26.5%, about $7M, had been converted to ETH by the attacker and was lost. Recovered funds were returned pro rata to affected LPs. This was an AMM-side loss for osETH liquidity providers, not a loss to vault stakers or to osETH backing.

Audits in the stakewise/v3-core repo: Halborn May 2023 (5 findings, 1 high), Halborn August 2023 (3 findings, 1 medium), Sigma Prime August 2023 (21 findings, 2 critical, 5 high, with LlamaRisk noting the majors involved the V2-to-V3 migration and were fixed), and a Sigma Prime report dated June 2024 whose scope this review did not read. A Hats Finance public audit competition was also run, and StakeWise says ABDK audited v4.0, a report not located. The Immunefi bug bounty has run since 2022-05-31 at a $200,000 maximum for critical and $50,000 for high; LlamaRisk flags that this deviates from the industry norm of 10% of funds at risk, and noted no bounty had been claimed as of mid-2024.

The exit

The docs say osETH redeems for ETH at its fair exchange rate: instant if enough unbonded ETH sits in the protocol, otherwise validators are exited, which rides the Ethereum consensus-layer exit queue, the same physics as every LST. Vaults v4.0 adds 0x02 execution-layer withdrawals, which StakeWise says shortens withdrawal times. A second, permissionless path exists at the contract: anyone can burn osETH against positions above 91.5% LTV, a hard-floor arbitrage against the worst-collateralized vaults.

Secondary liquidity is the weak layer. LlamaRisk’s June 2024 snapshot put the main venues at Curve osETH/rETH, Balancer osETH/WETH, and Uniswap V3 osETH/USDC, with liquidity deliberately spread across venues. The Balancer osETH/WETH pool was drained in the November 2025 exploit, and where that liquidity now sits was not verified; StakeWise’s site claims $80M of exit liquidity, a marketing figure this review could not confirm. Depth is far below stETH’s in any case.

Peg history is clean. LlamaRisk found deviations larger at launch when liquidity was thin, and by mid-2024 a small but consistent positive deviation of the secondary market rate, with daily volatility of 3.42% against ETH’s 3.37%. No osETH depeg event was found in any source reviewed, including through the November 2025 Balancer stress, though no continuous peg series through 2026 was located.

Concentration and dependencies

A query of StakeWise’s documented production subgraph on 2026-08-14 shows where the stake actually sits. The largest mainnet vaults by assets: the Genesis vault at 149,954 ETH and a 5% fee; an unnamed vault (0x4fef…47df) at 75,533 ETH and a 15% fee; another (0xe6d8…6c08) at 56,922 ETH and 5%; a fourth (0xb266…305b) at 24,319 ETH and 5%; the rest under 9,000 ETH each. The top 8 vaults sum to about 334,000 ETH, about $630M at $1,887 ETH, consistent with DefiLlama’s $702M once Gnosis and residual V2 are included. The Genesis vault holds roughly 40 to 45% of protocol stake and the top two vaults roughly 60%.

Inside Genesis, LlamaRisk’s June 2024 count had four operators across about 2,220 validators: StakeWise Labs 45.85%, CryptoManufaktur 27.79%, Deutsche Telekom 15.09%, Finoa 11.26%, meaning almost half of the validators were operated by StakeWise itself. In April 2025 StakeWise announced the Genesis vault, then nearly half a billion dollars, is moving to Obol distributed validators (blog.obol.org, 2025-04-02); completion status is unverified.

Dependencies. LlamaRisk measured 56.45% of all osETH sitting in the EigenLayer osETH strategy in June 2024 and flagged it as a threat to secondary liquidity; the current share is unverified. osETH is collateral on Aave V3 Ethereum, onboarded via a March 2024 ARFC and priced by Chainlink ETH/USD combined with the osETH rate under a CAPO cap, with LlamaRisk writing the supporting assessment; StakeWise also promotes borrowing against vault tokens on Morpho. Position sizes on those venues were not measured. The 11-entity oracle set is a trust dependency for the exchange rate, and LlamaRisk flagged that the RedStone price feed then available lacks Chainlink’s reliability, recommending a Chainlink feed before further collateral onboarding. There is no bridge dependency: staking and osETH are Ethereum-native, with Gnosis running a parallel osGNO deployment of about $1.1M, immaterial. On the corporate side, LlamaRisk flagged UAE free-zone incorporation with opaque ownership and an unclear VARA position on non-custodial staking. Fees across the top 8 live vaults run 0% to 15%, set by each vault admin, with Genesis at 5%; fee is part of the vault choice.

The comparison

Our Ethereum LST category review selected Lido and Rocket Pool, preferring the more decentralised validator set over marginally higher yield, and benched the rest, including StakeWise V2. V3 is a different design from V2, which is why it got this review. Scale at 2026-08-13: Lido $17.88B, Rocket Pool $0.99B, StakeWise $0.70B, within striking distance of Rocket Pool and about 4% of Lido.

What V3 offers that the selected pair do not. Permissionless vault creation with operator choice: a depositor picks the exact node operator, fee, and MEV policy, or requires a private allowlisted vault, while Lido stakers get the blended curated set plus modules and rETH holders get the blended permissionless set. Delegation for solo stakers and institutions: a client or chosen operator can run a dedicated vault, keep custody, and still mint a liquid token, which Lido’s CSM and Rocket Pool minipools do not give a depositor. Per-vault risk isolation: a bad operator burns their own vault’s depositors, not the whole protocol, with osETH insulated by overcollateralization, while stETH and rETH socialize slashing across the pool. Lido has announced its own stVaults in Lido V3, so the incumbent is copying the differentiator.

What StakeWise gives up. Liquidity depth and integration surface: osETH’s DEX liquidity is in the tens of millions against stETH’s presence in essentially every major venue, and its lending footprint is one Aave listing plus Morpho. Battle-testing at scale: Lido has operated at tens of billions since 2021 and Rocket Pool’s roughly 2,000 permissionless operators (per 2026 third-party reporting) have run through every market since 2021, against V3 code live since late November 2023 and v4.0 since October 2025. And realized validator distribution: the Genesis vault’s 40 to 45% of stake, with StakeWise Labs running about 46% of its validators at last verified count, is more concentrated than Rocket Pool’s spread, though the Obol DVT migration would improve it. The extra decision burden is also real: which vault is a live risk choice with vault-admin upgrade powers attached, exactly the question the universe entry posed.

Two structural notes an advisor should carry. A staker who mints osETH can access at most 90% of their stake as liquidity, the last 10% being locked buffer, and the oracle design pegs osETH accrual to the highest-LTV vault’s rate, so the effective yield on osETH differs structurally from the vault APY. And the DAO’s power to change per-vault LTV thresholds is a governance risk parallel to the Aave rsETH lesson, watch for LTV raises justified by growth; the 99.99% DAO-approved tier in the docs is exactly such a lever.

Open questions

The names and operators of the non-Genesis large vaults, 75,533 ETH at a 15% fee and 56,922 ETH at 5%, must be resolved before this file can reopen; the 15% fee suggests a managed or restaking product. Whether any vault operates at the 99.99% DAO-approved LTV tier, and the current DAO-set parameters generally, are unchecked against live state.

The slashing record between LlamaRisk’s June 2024 statement of none to its knowledge and today is unverified; no incident surfaced in searches, but absence of coverage is not absence of events. The current EigenLayer share of osETH (56.45% in June 2024) and per-venue secondary depth after the Balancer pool was drained are unmeasured, and the $80M liquidity claim is marketing. The completion status of the Genesis vault’s Obol DVT migration announced April 2025 is unknown.

The oracle voting threshold needs settling, since LlamaRisk states both 6 of 11 and 7 of 11 in the same report, as does whether the V3 oracle source code has since been opened. The June 2024 Sigma Prime report’s scope and findings, and the ABDK v4.0 report itself, were not read; StakeWise’s claim of a clean audit is its claim. Current Immunefi bounty terms (the 2024 figure is a $200k maximum) and the current multisig signer set (a 2024 snapshot of 4-of-7) are unverified. And the fall from $1.58B in October 2025 to $0.70B in August 2026 is undecomposed between the ETH price, about $1,887 at the review, and net ETH outflows.

Sources

The claims above trace to these. Where a number could not be independently verified, the thesis says so.

• StakeWise docs: osToken mechanics · primary · accessed 2026-08-15
  Supports: osETH overcollateralization, redemption and liquidation thresholds, oracle-controlled reward rate
• LlamaRisk collateral assessment: StakeWise osETH, 2024-06-25 · secondary · accessed 2026-08-15
  Supports: 2024 operator concentration snapshot, 2024 oracle and governance snapshot, historical liquidity assessment
• Chorus One guide to StakeWise V3 · secondary · accessed 2026-08-15
  Supports: vault mechanism, operator discretion, slashing allocation
• DefiLlama StakeWise V3 protocol data · secondary · accessed 2026-08-15
  Supports: protocol TVL history, chain allocation, scale comparison
• StakeWise documented production-subgraph endpoints, vault balances queried 2026-08-14 · primary · accessed 2026-08-14
  Supports: production subgraph endpoint, mainnet vault balances, vault fee concentration
• StakeWise V3 core and audit reports · primary · accessed 2026-08-15
  Supports: V3 source, Halborn reports, Sigma Prime reports, audit scope
• StakeWise mainnet deployment registry · primary · accessed 2026-08-15
  Supports: Genesis vault address, osETH contracts, oracle and factory addresses
• StakeWise Genesis Obol launch status, 2025-04-02 · primary · accessed 2026-08-15
  Supports: Obol launch, 10% initial validator share, planned expansion
• StakeWise DAO recovered-funds proposal, 2025-11-25 · primary · accessed 2026-08-15
  Supports: 6851 osETH affected, 5041 osETH recovered, LP distribution method

Inherited controls

The verdict above grades the protocol layer. Every position also inherits the asset it holds and the chain it settles on. The least safe layer sets the position’s grade, and the position table names which one that is.