Skip to content

Guide · Staking

How Ethereum staking works: validators, withdrawals and slashing

How Ethereum validators actually work: the activation queue, what specifically triggers slashing, the two withdrawal credential types, and how the exit and withdrawal sweep processes real balances. Not financial advice.

By Etherelysium Editorial TeamUpdated 18 November 20267 min read

How Ethereum staking works: validators, withdrawals and slashing
Share X LinkedIn

Running an Ethereum validator means depositing 32 ETH, running the required software, and taking on two jobs the protocol pays you for: proposing blocks when chosen, and attesting — voting on which block is the correct head of the chain — once every epoch. It also means accepting two real downside risks in exchange for that reward: being penalized for simply going offline, and being slashed for specific, provable misbehavior.

Becoming a validator

A validator needs 32 ETH deposited into the protocol’s deposit contract, plus three pieces of running software: an execution client, a consensus client, and a validator client. New validators don’t join instantly — they wait in an activation queue that limits how fast the overall validator set can grow, which is a deliberate throttle on the network’s own rate of change rather than a processing delay.

Once active, a validator’s core job repeats every epoch: a randomly selected committee member proposes a block when it is that validator’s turn (once per roughly 6.4-minute epoch across the whole set, statistically), and every active validator submits one attestation per epoch voting for what it believes is the canonical chain. The chain’s fork-choice rule tallies these votes to converge on a single view of the correct chain, as described in the companion guide on finality and reorgs.

What gets you slashed

Slashing targets specific, provable bad behavior, not honest mistakes or bad luck. The two slashable offenses are proposing two different blocks for the same slot (equivocating) and submitting contradictory attestations that violate the protocol’s voting rules. Both are the kind of action a validator client should never produce under correct operation — they typically happen from misconfigured redundant setups (running the same validator keys on two machines at once) rather than deliberate attacks by a solo honest operator.

The penalty for slashing scales with how many validators are slashed around the same time — the correlation penalty. A single, isolated slashing event costs under 0.1% of the validator’s stake. A mass slashing event, where many validators are slashed simultaneously, can destroy the entire staked balance of each one involved, because the protocol reads a large number of simultaneous slashings as a stronger signal of a coordinated attack rather than independent operator error, and punishes accordingly.

Simply going offline is a separate, much milder penalty: missed attestations cost small, ongoing rewards rather than a slashing-level loss, since non-participation does not threaten the chain’s safety the way equivocation does — it only costs the inactive validator the rewards it would otherwise have earned.

Withdrawal credentials: two types

A validator’s withdrawal credential determines where its ETH can go and how its balance behaves above 32 ETH. Type 1 (legacy) credentials cap effective balance at 32 ETH and automatically sweep out any excess rewards periodically at no cost to the operator. Type 2 (compounding) credentials can hold an effective balance up to 2048 ETH, letting rewards compound inside the validator instead of being swept out — but a Type 2 validator’s own operator must actively trigger a partial withdrawal to pull out excess below that higher ceiling, and doing so costs gas. Converting from Type 1 to Type 2, or setting a withdrawal address in the first place, is a one-way, irreversible action.

Exiting and getting your ETH back

A voluntary exit can be signed directly with the validator’s own key through the validator client, at no gas cost, or triggered from the execution layer via the withdrawal address, which does cost gas. Once an exited validator is flagged “withdrawable,” its full remaining balance goes out in the next sweep that reaches it.

Withdrawals — both the automatic partial sweeps above 32 ETH and full exits — are processed through a sweep that starts at validator index zero and moves forward, wrapping around after the last index, building a queue of up to 16 withdrawals per block. At full throughput the network processes roughly 115,200 withdrawals a day; a very large backlog of validators all exiting at once can therefore still take days to fully clear, since the sweep mechanism itself, not any single operator’s choice, sets the pace.

Forced exits run on their own clock

A validator pushed out involuntarily — for prolonged inactivity below a safety threshold, for example — goes through roughly a 36-day process: an immediate small penalty capped at 0.5 ETH on day one, the correlation penalty applying around day eighteen if slashing is involved, and ejection from the network by around day thirty-six.

Why 32 ETH specifically

The 32 ETH deposit size is a deliberate balance between two competing goals: low enough that a meaningful number of independent participants can realistically run a validator, and high enough that the total number of validators — and therefore the number of attestations the network has to process and the committee math has to account for every epoch — stays within what the consensus layer can handle efficiently as the validator set scales into the hundreds of thousands.

Committees and randomness

Rather than having every active validator vote on every single slot, the protocol splits the validator set into committees, each assigned to a specific slot within the epoch, so that every validator attests exactly once per epoch while the chain still receives a steady stream of attestations slot by slot. Which validator proposes which block, and which committee attests to it, is determined by a source of on-chain randomness the protocol generates itself, specifically so that no participant can predict or manipulate their own selection in advance.

This randomness is also why running more validators does not let an operator game proposer selection — each additional validator simply gets its own independent, equally random chance of being chosen, rather than compounding into a disproportionate advantage for larger operators relative to their actual share of total staked ETH.

Each of these mechanics — the queue, the committees, the randomness — exists to keep the validator set’s growth and its day-to-day operation predictable and resistant to manipulation, rather than to make staking itself faster or slower for any individual operator.

For a reader deciding whether to run a validator directly or use a pooled or liquid staking option instead, the honest tradeoff is operational burden and these specific slashing and queue mechanics on one side, versus the intermediary and smart-contract risks covered in the companion guide on liquid staking and restaking on the other — not a simple question of which pays more, since both ultimately draw on the same underlying consensus-layer rewards.

Either path still ultimately depends on the same base protocol described above functioning correctly.

Risk note

Running a validator carries real operational risk: a misconfigured redundant setup is the most common real-world cause of slashing, not a market event. Liquid staking, covered in the companion guide, exists specifically for people who want staking rewards without taking on this operational burden directly — at the cost of a different set of risks.

Not financial advice. Nothing on this page is a recommendation to buy, sell, stake, bridge, or avoid any asset or protocol.

FAQ

Can I get slashed just for my node going offline?

No. Downtime triggers a much smaller penalty on missed rewards, not slashing. Slashing specifically targets provable misbehavior: proposing two blocks for one slot, or submitting contradictory attestations.

How long does it take to fully exit a validator?

A voluntary exit followed by a full withdrawal depends on the current exit and withdrawal sweep queue, which processes roughly 115,200 withdrawals per day at full throughput; an involuntary forced exit runs on its own roughly 36-day schedule.

What’s the difference between the two withdrawal credential types?

Type 1 (legacy) caps effective balance at 32 ETH and auto-sweeps excess rewards for free. Type 2 (compounding) can hold up to 2048 ETH and lets rewards compound, but pulling out excess below that ceiling requires the operator to actively trigger a partial withdrawal and pay gas.

Why is a mass slashing event so much worse than a single one?

The penalty includes a correlation component that scales up sharply when many validators are slashed around the same time, since the protocol reads that pattern as a stronger signal of coordinated attack rather than independent error.

Frequently asked questions

Can I get slashed just for my node going offline?
No. Downtime triggers a much smaller penalty on missed rewards, not slashing. Slashing specifically targets provable misbehavior: proposing two blocks for one slot, or submitting contradictory attestations.
How long does it take to fully exit a validator?
A voluntary exit followed by a full withdrawal depends on the current exit and withdrawal sweep queue, which processes roughly 115,200 withdrawals per day at full throughput; an involuntary forced exit runs on its own roughly 36-day schedule.
What's the difference between the two withdrawal credential types?
Type 1 (legacy) caps effective balance at 32 ETH and auto-sweeps excess rewards for free. Type 2 (compounding) can hold up to 2048 ETH and lets rewards compound, but pulling out excess below that ceiling requires the operator to actively trigger a partial withdrawal and pay gas.
Why is a mass slashing event so much worse than a single one?
The penalty includes a correlation component that scales up sharply when many validators are slashed around the same time, since the protocol reads that pattern as a stronger signal of coordinated attack rather than independent error.

Educational content, human-edited and possibly AI-assisted. It describes how systems work and is not financial, legal or tax advice. Report an error via our corrections page.

More in Staking

Related articles

See all