DeFi
What restaking is, and what each layer adds
Restaking takes ether that already secures Ethereum and makes it answer to other services as well, which pay for that security and can destroy part of it when their rules are broken. EigenLayer, which DefiLlama lists as EigenCloud, was the largest protocol it filed under Restaking as of 22 November 2026. Eigen Labs announced on 17 April 2025 that slashing was going live on mainnet that day, and stake becomes slashable only when an operator allocates it to a particular service. Liquid restaking tokens then wrap the position again. Each layer adds a failure mode of its own, and a holder at the top of the stack carries every one of them.
The idea: reuse stake that is already at risk
A validator’s stake is collateral. Ethereum destroys part of it for provable misbehaviour, and that threat is what makes the validator’s signature worth trusting. ethereum.org’s restaking page describes restaking as using that already-staked ether to secure other decentralised services, which pay additional rewards for it, while the ether is put at more risk. The services are usually called actively validated services, or AVSs; EigenLayer’s documentation now expands the same initials as Autonomous Verifiable Services. ethereum.org gives price oracles, token bridges and data systems as examples.
There are two ways in. A native restaker runs Ethereum validators and points their withdrawal credentials at EigenLayer’s contracts. A liquid restaker deposits a token, typically a liquid staking token, the claim on pooled stake described in what liquid staking is. ethereum.org draws one boundary worth keeping: lending out a liquid staking token in DeFi is not restaking, because restaking means securing a service, not earning on the token.
The argument for the design, as EigenLayer’s overview makes it, is that a new service need not raise its own capital for security and recruit its own operators from nothing. The argument for caution came early. Vitalik Buterin’s post Don’t overload Ethereum’s consensus, of 21 May 2023, treats reusing validator stake as low-risk when any losses stay with the validators and users who opted in, and high-risk when an application expects Ethereum’s social consensus, through a fork or a reorganisation, to rescue it if it fails.
How EigenLayer’s model works
EigenLayer accepts native ETH, liquid staking tokens, its own EIGEN token and other ERC-20 tokens into its contracts, according to its overview. Its glossary defines the roles around them.
Stakers deposit the assets. Each one delegates its whole restaked balance to a single operator at a time, an all-or-nothing choice, and a native restaker holds its position in an EigenPod, a contract each address can deploy once and whose ownership cannot be transferred, according to the restaking overview. Operators run the services’ software. An operator cannot take the tokens delegated to it, but it can make them slashable.
The AVSs sit on the other side. Each organises operators into operator sets, gives them tasks, pays rewards through a contract called the RewardsCoordinator, and writes the conditions under which it slashes. Allocation is what connects an operator to them: the operator commits proportions of its delegated stake to particular operator sets, and stake committed this way becomes Unique Stake, slashable by the AVS that created that set and by no other.
Rewards arrive from two places. AVSs pay in an ERC-20 token that each rewards submission names, under ELIP-001, and the protocol mints EIGEN to stakers and operators as what it calls programmatic incentives. The second is token issuance, paid for by the token’s other holders in the sense the tokenomics piece uses, and where crypto yield comes from covers EigenLayer’s points programme and the airdrop that followed it.
Delays are built in. ELIP-002, the proposal that introduced slashing, lists the mainnet parameters, each set through governance: about 17.5 days before a change to an operator’s allocation delay takes effect, so that its stakers can react; about 14 days before a deallocation takes effect; and about 14 days between a staker queuing a withdrawal and being able to complete it. The withdrawal can still be slashed during those 14 days, and any slashing is applied when it completes. The allocation delay itself is the operator’s to set, and EigenLayer’s safety delays reference says the protocol puts no constraint on it. ELIP-016 records that a number of operators, public ones included, had set it to zero, which allows instantaneous changes to their risk profile. A native restaker leaving altogether first waits out Ethereum’s exit queue, then a fixed 256 epochs (about 27 hours) before the ether becomes withdrawable, then a validator sweep that EigenLayer’s withdrawal page puts at up to about nine days.
What slashing means here, and when it was switched on
On Ethereum, slashing punishes a short list of offences the protocol can prove, such as signing two blocks for one slot. On EigenLayer, each AVS writes its own. EigenLayer’s slashing overview calls the function maximally flexible: an AVS may slash for any reason, and the reason does not have to be provable on-chain. AVSs are encouraged to make their slashing legible, and operators are responsible for understanding the conditions before they take them on. An AVS can slash an operator only up to the Unique Stake allocated to its operator set, only while the operator is registered there or within the delay after it leaves, and only in a strategy that set includes.
Slashing came after rewards. ELIP-002 describes its upgrade as the one that gives AVSs the ability to slash. A testnet slashing release of the contracts was published on 19 December 2024 and the mainnet release on 3 April 2025, and Eigen Labs announced on 17 April 2025 that slashing was going live on mainnet later that day, once its Protocol Council completed the final steps. The announcement calls it opt-in: a staker’s funds are exposed only if their operator joins a slashable operator set. The documentation states the consequence from the staker’s side. Once an operator opts in and allocates, stake that a staker delegated earlier becomes slashable, and stakers are responsible for continuing to monitor what their operator allocates.
ELIP-002 says a slashing reduces, in proportion, the funds of every staker delegated to that operator in the affected strategies, including funds in withdrawals still inside the 14-day delay. Slashed funds were at first only burned. ELIP-006 added redistributable operator sets, which the documentation says are available on mainnet: slashed funds go to a recipient address fixed when the set is created. The restaking overview is plain about what that changes. There is a larger incentive to slash when funds can be redistributed instead of burned; if an AVS’s governance or slashing logic is corrupted, an attacker may be able to drain delegated funds; and a compromised operator could set up an AVS of its own to take them. Native ETH and EIGEN are excluded from redistribution, and slashed native ETH stays locked in EigenPods permanently. A June 2026 release of the contracts, v1.13.0, added the slash resolution delay that ELIP-016 proposed: slashed shares wait 50,400 blocks, about seven days, before they can be burned or redistributed, and a new pause flag lets the pauser multisig halt that step, which ELIP-016 says is for security incidents.
Liquid restaking tokens on top
A liquid restaking token does for a restaked position what a liquid staking token does for a staked one. A protocol takes deposits, restakes them through EigenLayer, picks the operators, and issues a token that can be transferred. ethereum.org uses eETH as its example, but ether.fi, which issues it, now says it no longer restakes ETH through EigenLayer and its FAQ describes eETH as a liquid staking token. Two other tokens show the arrangement in more detail.
Renzo’s ezETH is minted against deposits of ETH or stETH, according to Renzo’s ezETH page. It is reward-bearing: the balance stays fixed and its value against the underlying assets rises. Renzo charges 10% of the rewards generated by restaking, split equally between its treasury and its node operators. Native ether waits in a Renzo contract, the DepositQueue, until 32 ETH has accumulated, then goes to a validator whose withdrawal credentials point at an EigenPod, and Renzo’s contracts are upgradeable through proxies, with upgrades gated by a protocol multisig (how Renzo works). Its withdrawals page lists three delays: Ethereum’s exit, EigenLayer’s 14-day withdrawal delay and a three-day buffer delay of Renzo’s own. It expects a wait of seven days when its buffer holds assets and 10 to 15 days when it does not, without saying how the delays overlap to produce those ranges.
Kelp’s rsETH is reward-bearing too. Its rate comes from an oracle contract that values every asset the protocol holds in ETH and divides the total by the rsETH supply, so no single swap can move it, according to Kelp’s page on the rate. No single update may currently move the rate by more than 1%, a fall beyond the limits pauses deposits, withdrawals and further updates, and the limits can be changed by a 6-of-11 admin multisig. On layer-2 networks the rate arrives as a cross-chain message. A January 2026 upgrade, recorded on Kelp’s upgrades page, gave the rsETH contract a 24-hour hold and recovery function per address, so that Kelp can freeze and sweep tokens held by compromised addresses, while partners it describes as major DeFi protocols, exchanges and market makers stay exempt.
The token also hides a choice. Renzo’s documentation counts the combinations of services a position could secure and finds 32,767 with 15 AVSs. A holder of a liquid restaking token has handed that choice, and the choice of operator, to the protocol that issued it.
As of 22 November 2026, per DefiLlama, the protocols it files under Liquid Restaking with at least $1M each held $793.4M, across 11 protocols listed on the liquid restaking category page. Kelp accounted for $603.5M of it, 76.1% (computed), and Renzo for $75.3M. The category is DefiLlama’s label, and a product can change what it is: ether.fi Stake, whose eETH ethereum.org still gives as its liquid restaking example, is filed under Liquid Staking, in line with ether.fi’s own account that eETH no longer restakes. EigenLayer is listed as EigenCloud, the name DefiLlama now gives it, under a separate Restaking category, with $4.3B of its own. The same ether can appear in more than one of these figures, the layering what TVL measures works through: a liquid staking protocol counts the ether it staked, and a restaking protocol that holds the token counts it again.
How the risks stack
Follow one unit of ether up a full stack. It is staked through a liquid staking protocol; the token goes to a liquid restaking protocol, which restakes it on EigenLayer through operators serving several AVSs; and the holder posts the resulting token as collateral in a lending market. Each layer is a separate place it can be lost.
- The validator. Ethereum’s slashing and penalties, and Ethereum’s exit queue.
- The liquid staking token. Its contracts, governance and oracle, its operators, and its market price.
- EigenLayer’s contracts. Code that can fail, and delays and pausers that governance sets.
- The operator. One per staker, making allocations the staker has to watch; a compromised operator is one of the documented routes to theft.
- The AVSs. Slashing conditions each writes for itself, which need not be provable on-chain, and governance that can redistribute what it slashes if it is corrupted.
- The liquid restaking token. Another set of upgradeable contracts and multisigs, another rate, another withdrawal queue, and a market price that can sit below the rate, as any liquid staking token’s can.
- The collateral. A lending market that values the token through its own oracle and liquidates when that value falls, as how DeFi lending works explains.
Two features make the stack worse than a list. The first is correlation. ethereum.org warns of chain reactions: a restaker slashed while securing several AVSs leaves less security behind for the others. Unique Stake is EigenLayer’s answer on the services’ side, since each allocation is slashable by one operator set only, but a staker’s position still adds up every allocation its operator makes. The second is time. Each layer’s delay has to clear before ether comes back, which is why Renzo’s estimate with an empty buffer runs to 15 days, and a holder who needs to leave faster sells the token for whatever the market pays.
Then there is the reward. Restaking pays for extra risk, and ethereum.org’s advice, for any figure meaningfully above the network’s own staking rate, is to ask exactly where the extra comes from. Some of it is AVS payments, some is emission of EIGEN or of a restaking protocol’s own token, and some has been points with no guaranteed conversion into anything. None of it is a property of the ether underneath.
What to ask of a restaking position
- Which operator, and which operator sets? Including whether any are redistributable, which ELIP-006 makes readable on-chain for each set.
- Who can upgrade or pause each contract in the stack? A multisig at every layer is a set of keys at every layer.
- How long does a full exit take? The delays add up; the shortest one is not the answer.
- Where does the token’s rate come from, and who can change its limits?
- Which rewards are paid in which token? The staked asset, an AVS’s own token and protocol emissions each carry different risks.