At 15:53 UTC on 11 October 2023, Lido’s monitoring flagged its first two cases of Ethereum slashing. By 16:05 the count had stopped at 20. All of them were run by Launchnodes, one of Lido’s node operators, which had lost connectivity in its main data centre and switched to a standby site while the first one was still partly alive. For a few minutes the same validator keys were signing in two places at once, and the network saw two different votes from the same validators. No attacker was involved and Ethereum had no bug: the code punished exactly what it was written to punish, according to Lido’s post-mortem.
That is Ethereum slashing in practice. Each of the 20 validators lost 1 ETH on the spot, the rule in force at the time. Lido put the projected total cost, penalties and missed rewards included, at 28.677 ETH. Launchnodes sent 25.663 ETH in compensation the next morning and promised to cover the rest until the slashed validators had left the network, so stETH holders saw no cut in their rewards that day.
A validator is a job with a pay slip and a penalty clause, and both fit in a few lines of the protocol’s specification. This lesson reads those lines: where the pay comes from and why it shrinks as more ETH is staked, what being offline costs, which behaviours are slashed and how the bill is calculated since the Pectra upgrade, what real incidents looked like, and how Cosmos, Polkadot, Solana and Cardano handle the same job. The validator lifecycle itself (deposits, queues, attestations, committees) is the subject of the previous lesson, our guide to proof of stake in depth: here it is taken as known.
Three outcomes are possible for a validator in any epoch of 6.4 minutes. It does its duties and is paid. It misses them and is fined a small amount. Or it signs two messages that contradict each other, and it is slashed: fined, forced out and kept waiting for weeks before it can withdraw. The first two happen to every operator. The third is rare and almost always self-inflicted.
Where a validator’s pay comes from
An Ethereum validator has two sources of income. Consensus-layer rewards are new ETH minted by the protocol for voting and proposing, credited to the validator’s balance on the Beacon Chain. Execution-layer rewards are the priority fees (tips) that users add to their transactions, plus any MEV payment from a block builder, and they go to the address the validator names when it proposes a block. The base fee of each transaction is burned under EIP-1559, so it pays nobody.
Five duties and one formula for consensus rewards
The consensus-layer pay is set by a single number, the base reward. Per epoch, each unit of 1 ETH of effective balance earns 64 gwei multiplied by 1 billion and divided by the square root of all active stake counted in gwei. A validator’s base reward is that figure times its effective balance in whole ETH. The functions are in the Altair beacon-chain specification, still in force with later changes layered on top.
The base reward is then split between five duties, with weights that add up to 64:
| Duty | Weight (out of 64) | Share of the maximum reward | Penalty if missed |
|---|---|---|---|
| Timely source vote (attestation) | 14 | 21.9% | Yes, same amount |
| Timely target vote (attestation) | 26 | 40.6% | Yes, same amount |
| Timely head vote (attestation) | 14 | 21.9% | No, only the reward is lost |
| Sync committee participation | 2 | 3.1% | Yes, for members who miss |
| Block proposal | 8 | 12.5% | No, only the reward is lost |
Attestations, the three votes every active validator casts once per epoch, account for 54 of the 64 parts, 84.4% of the total. One detail matters for anyone who runs a node: the attestation rewards are scaled by how much of the network voted correctly. When many validators miss, everyone who did vote earns a little less.
A worked example with today’s numbers. On 8 November 2026 at 09:48 UTC, a public beacon node reported 814,255 active validators with 44,379,824 ETH of total effective balance, by our count of beacon-chain data. The square root of that balance in gwei is about 210.7 million, so the base reward per ETH per epoch is 303 gwei (the specification rounds down). A 32 ETH validator therefore earns up to 9,696 gwei per epoch. With 225 epochs a day, that is close to 0.80 ETH a year, or 2.5% on 32 ETH, provided it performs every duty and counting the consensus layer only.
Two public dashboards gave the same order of magnitude that morning: validatorqueue.com showed a staking APR of 2.43% with 44.4 million ETH staked, and ethereum.org’s staking page showed about 2.5%. Neither page says in so many words whether execution-layer income is included, so treat 2.4% to 2.5% as the consensus-layer rate as of 8 November 2026.
Why the rate falls when more ETH is staked
The square root in the denominator is a design choice. Total issuance grows with the square root of the stake, so the yield per validator falls as stake rises: four times more ETH staked means half the rate. When few people stake, the protocol pays more to attract them, and when many do, each one gets less. The table applies the same formula to different amounts of staked ETH, full participation assumed:
| Total ETH staked | Consensus-layer yield per year | Note |
|---|---|---|
| 10 million | 5.26% | |
| 20 million | 3.72% | |
| 30 million | 3.04% | |
| 44.4 million | 2.50% | level of 8 November 2026 |
| 64 million | 2.08% | |
| 100 million | 1.66% | more than three-quarters of today’s supply |
At today’s stake the protocol mints about 1.1 million ETH a year for validators, by our arithmetic (0.80 ETH times 44.4 million divided by 32). That 2.5% is a rate on staked ETH, not on the whole supply, and it is a nominal figure: rewards are paid in new ETH, so the yield in dollars depends on the price, and part of the yield is offset by the dilution of everyone who does not stake. The difference between the quoted rate and what compounding gives is covered in our lesson on APR versus APY.
Block proposals and sync committees work like a lottery
The averages above hide a lot of variance. Ethereum has 7,200 slots a day, one every 12 seconds, and each slot has one proposer picked at random, weighted by effective balance. With 814,255 active validators, a 32 ETH validator can expect about 3.2 proposals a year. Some will get none in twelve months, others five or six, and the 12.5% share of consensus rewards tied to proposals arrives in those few slots.
The sync committee is rarer still. It has 512 members, redrawn every 256 epochs (about 27.3 hours). The chance that a given validator is picked for one period is about 0.063%, which adds up to roughly one chance in five or six over a year. Members earn an extra reward for each slot they sign during their period, and lose about the same amount for each slot they miss.
Tips and MEV arrive in lumps
Execution-layer income follows the proposals. A validator earns tips and MEV only in the slots where it proposes a block, so a solo validator with three proposals a year collects that income three times, in amounts that depend on how busy the network is at that moment. Published estimates of its share of total validator income vary widely with the period and the method, so no single figure is given here. Most validators take blocks from outside builders through MEV-Boost, covered with its risks in our guide to MEV protection and proposer-builder separation. Where MEV comes from in consensus is the subject of a later lesson of this path, MEV explained.
What 2,048 ETH validators change
Compounding validators, the “0x02” type that Pectra introduced on 7 May 2025 and that the previous lesson describes, can hold up to 2,048 ETH of effective balance and keep their rewards on the balance. On 8 November 2026, 24,498 validators used 0x02 credentials, 3.0% of the active set, by our count of beacon-chain data. Everything in this lesson scales with effective balance: a 2,048 ETH validator earns 64 times more than a 32 ETH one, is picked to propose 64 times as often, and pays 64 times more when it is penalised or slashed. That last point is why bonds grow with validator size: when Lido opened a permissionless route to 2,048 ETH validators on 23 October 2026, it asked operators for a 32 ETH bond on the first one, as reported in our news on Lido’s 0x02 module.
Going offline costs about what being online pays
Missing attestations does not get a validator slashed. It is fined, and only for two of the three votes: the source and target votes cost exactly what they would have paid, while a missed head vote only forfeits its reward. That rule is visible in the function that computes attestation deltas in the Altair specification.
In numbers, at today’s stake, a 32 ETH validator that is offline loses about 6,100 gwei per epoch in penalties, or about 0.0014 ETH a day, on top of the roughly 0.0022 ETH a day it would have earned. Compared with staying online, a day offline therefore costs about 1.6 days of income. A machine that is down for a weekend loses a few hundredths of an ETH and recovers it in a few days once back. On attestations alone, the break-even sits at about 43% uptime (40 parts lost against 54 earned), so a validator that is online most of the time ends the year with a gain.
The inactivity leak when finality stops
There is one case where being offline becomes expensive: when more than a third of the stake is offline and the chain stops finalising. After four epochs without finality, the protocol starts the inactivity leak, which drains the offline validators until those still voting again hold two-thirds of the stake. The mechanism and the May 2023 episode when mainnet lost finality are told in our lesson on finality and reorgs. What it lacks is the arithmetic.
During a leak, each offline epoch adds 4 points to a validator’s inactivity score, and its penalty per epoch is its effective balance times that score divided by 4 times 224. The loss therefore grows with the square of the time spent offline. By our arithmetic from the specification, for a 32 ETH validator that stays offline throughout a leak:
| Time offline during a leak | Approximate inactivity penalty |
|---|---|
| 1 day (225 epochs) | 0.05 ETH |
| 4.4 days | 0.93 ETH |
| 9.1 days | 4 ETH |
| 18.2 days (4,096 epochs) | 16 ETH, about half the balance |
Once its effective balance falls to 16 ETH, a validator is ejected from the active set, which caps the damage close to that level. The figures are approximate: the effective balance drops in steps during the leak, which slows the losses a little. They show why a leak is the protocol’s last resort and why a short outage of a few validators costs almost nothing.
Ethereum slashing: the offences and the three charges
Two signatures that cannot both be honest
Slashing punishes signatures, not silence. The specification knows three offences:
- A proposer signs two different blocks for the same slot.
- An attester signs two different votes for the same target epoch, a double vote.
- An attester signs a vote whose source and target surround those of another of its votes, a surround vote.
Each one is what an attack on the chain would look like: voting for two competing histories so that both can be finalised, or rewriting a vote to support a different past. An honest validator never has a reason to do it, so the protocol treats two conflicting signatures as proof of guilt, whatever the intention. In practice, almost every slashing on mainnet so far comes from the same operational error as Launchnodes: one set of keys running on two machines at the same time.
Any validator that sees the two signed messages can package them into a proof. A block proposer includes it, and the slashing takes effect in that block. A block can carry up to 16 proposer slashings and, since Pectra, one attester slashing, but a single attester slashing can cover many validators that signed the same pair of votes.
The initial penalty since Pectra: 1/4096 of the balance
The first charge is immediate and has been cut sharply over the years. It is a fixed fraction of effective balance, defined by a constant that each upgrade has changed. Altair and Bellatrix raised it in two steps towards its full value, as the ethereum.org history of upgrades notes, and Electra cut it:
| Upgrade (activation) | Initial penalty | On a 32 ETH validator |
|---|---|---|
| Phase 0 (1 December 2020) | 1/128 of effective balance | 0.25 ETH |
| Altair (27 October 2021) | 1/64 | 0.5 ETH |
| Bellatrix (6 September 2022) | 1/32 | 1 ETH |
| Electra, part of Pectra (7 May 2025) | 1/4096 | about 0.0078 ETH |
The Pectra value is set by the constant MIN_SLASHING_PENALTY_QUOTIENT_ELECTRA, equal to 4,096 in the mainnet presets and confirmed on a live mainnet node on 8 November 2026. Many explanations online, including well-known pages written before 2025, still give the old figure of 1/32 or “at most 1 ETH”. That is outdated: the Launchnodes validators paid 1 ETH each because they were slashed in 2023. A 2,048 ETH validator slashed today pays 0.5 ETH up front.
The correlation penalty: cheap alone, ruinous in a crowd
The second charge is the one that matters. About 18 days after a slashing, the protocol looks at all the effective balance slashed in a window of 8,192 epochs (about 36 days) centred on that moment, multiplies it by 3, and takes that share of every slashed validator’s balance. The rule is in the process_slashings function of the Electra specification: if 1% of all staked ETH was slashed in that window, each slashed validator loses 3% of its balance; if a third or more was slashed, it loses everything.
| Share of all staked ETH slashed in the window | Correlation penalty on a 32 ETH validator |
|---|---|
| One validator alone (32 ETH out of 44.4 million) | about 0.00007 ETH |
| 1% | 0.96 ETH (3%) |
| 5% | 4.8 ETH (15%) |
| 10% | 9.6 ETH (30%) |
| 20% | 19.2 ETH (60%) |
| One third or more | 32 ETH, the whole balance |
The design targets the case Ethereum fears most. An attack on finality needs at least a third of the stake to sign conflicting votes, and that is exactly where the penalty reaches 100%. An isolated operator error costs almost nothing extra. The flip side is that the risk is shared: a bug in a client used by a large share of validators, or a staking provider running many keys on the same faulty setup, could push slashed balances high enough for the correlation penalty to bite every one of them.
Forced exit and 36 days without pay
The third charge is time. A slashed validator is put in the exit queue at once and cannot withdraw until 8,192 epochs, about 36.4 days, have passed. During that period it earns nothing and is still charged the source and target penalties of an offline validator. By our arithmetic from the specification and the stake of 8 November 2026, that adds up to about 0.05 ETH on 32 ETH, plus about 0.08 ETH of rewards it would otherwise have earned.
| Scenario (32 ETH validator, stake of 8 November 2026) | Approximate cost |
|---|---|
| Lone slashing, all charges | about 0.058 ETH, plus about 0.08 ETH of forgone rewards and a forced exit |
| Same slashing in 2023, before Pectra | about 1.05 ETH plus forgone rewards |
| Slashed with 1% of all stake | about 1 ETH |
| Slashed with a third of all stake | the whole 32 ETH |
| 2,048 ETH validator slashed alone | about 0.5 ETH initial, plus penalties scaled by 64 |
Who gets paid for catching it
The proposer of the block that includes the proof receives a reward of 1/4096 of the slashed validator’s effective balance, about 0.0078 ETH for a 32 ETH validator. Before Pectra the quotient was 512, so the reward was eight times larger. The specification still provides for a separate whistleblower, but block processing leaves it to the proposer, which in practice collects the whole amount.
Slashing incidents so far: operators’ mistakes, not attacks
The three best-documented cases on mainnet share the same cause.
| Date | Operator | Validators slashed | Cause, as stated by the operator or its partner | Cost |
|---|---|---|---|---|
| Early February 2021 (reported 4 February) | Staked | 75 | persistence of Prysm’s slashing-protection database switched off to improve signing | about 18 ETH (about $30,000) per its CEO; customers compensated |
| 11 October 2023 | Launchnodes, for Lido | 20 | failover to a standby data centre while the first still ran, so duplicate attestations | 20 ETH initial; 28.677 ETH projected in total; 25.663 ETH paid by Launchnodes |
| 10 September 2025 | One cluster owned by Ankr, on SSV Network | 39 (plus 1 unrelated, cause unconfirmed) | a validator instance run in parallel outside SSV during maintenance | not stated in the post-mortem |
Staked, a staking provider, told The Block in February 2021 that it had turned off the persistence of the Prysm client’s slashing-protection database while trying to improve its validators’ signing rate. Its chief executive, Tim Ogilvie, put the penalty at about 18 ETH, around $30,000 at the time, and said customers would be fully compensated, according to The Block’s report.
The Launchnodes case, told above, followed a detailed timeline: offline alerts at 15:41 UTC, the first slashing alerts at 15:53, all 20 confirmed at 16:05. Lido’s report lists the fixes: the slashing database of its remote signer switched on, the old nodes and signer shut down, and a new signer with slashing protection before the remaining validators came back in batches. A later update to the same report says the 20 slashed validators became withdrawable on 16 November 2023, 36 days after the event, and stopped accruing penalties.
The September 2025 case is often described as an SSV failure. The post-mortem published by SSV Labs says otherwise. On 10 September 2025, one validator was slashed at 11:51 UTC, and a cluster of 39 validators belonging to Ankr from 13:17 UTC. SSV’s node logs showed a single signature from its side in both cases. Ankr confirmed that afternoon that a maintenance mistake had started a second copy of its validators outside the SSV network, a key-management error on its own side. SSV gave no figure for the ETH lost.
The overall count stays small. On 8 November 2026, 594 validators carried the slashed flag out of 2,403,329 validators ever registered on the Beacon Chain since its launch on 1 December 2020, about 0.025%, by our count of beacon-chain data. None of them is still active. Explorers such as beaconcha.in show each case with its date and the block that included the proof.
The defences follow from the cause. Every serious validator setup keeps a slashing-protection database that refuses to sign a message conflicting with one already signed; EIP-3076 defines a format to move that history between clients. The other rule is organisational: one key runs in one place, and a standby machine is started only after the first one is known to be dead, never “just in case”. Downtime costs little; a hurried failover can cost much more.
Running a validator or delegating: who pays for a slashing
A solo validator needs 32 ETH, a machine that stays online, an execution client and a consensus client, and the discipline described above. It also needs patience: on 8 November 2026 about 944,812 ETH was waiting in the entry queue, a wait of about 16 days and 21 hours according to validatorqueue.com, and a new validator earns nothing while it waits. Leaving was faster that day, about 1 day and 3 hours in the exit queue, followed by a withdrawal sweep of about 7 days. A slashed solo staker bears every charge alone.
Most ETH holders do not run their own node. With staking-as-a-service, the holder keeps the deposit but an operator runs the keys, and the contract decides who pays for mistakes. In pooled and proof of stake coin with a protocol that stakes it for you and issues a transferable token (such as Lido stETH or rETH) representing the deposit plus rewards, which you can use or sell while the coins stay staked.">liquid staking, the smart contract holding reserves of two or more tokens, supplied by users, that a decentralized exchange uses to execute trades automatically at a price set by a formula.">pool spreads the stake over many operators. Lido, the largest, asks the operators of its permissionless modules to post an ETH bond that absorbs losses they cause, and the Launchnodes case shows the other route: the operator paid out of pocket. How liquid staking tokens such as stETH and rETH work, and what can go wrong with them, is in our guide to liquid staking tokens.
On chains with native delegation the answer is written into the protocol. On the Cosmos Hub, a holder who delegates ATOM to a validator shares that validator’s penalties pro rata. On Cardano, delegated ADA is never at risk.
Other chains: the same job, different rules
| Chain | Reward source | Slashing | Can delegators lose principal? | Exit or unbonding delay |
|---|---|---|---|---|
| Ethereum | New ETH (about 2.5% a year on stake) plus tips and MEV | Yes: 1/4096 initial, correlation penalty, forced exit | Not natively; through a pool or operator, depending on its terms | Exit queue (about 1 day on 8 Nov 2026) plus sweep; slashed: about 36 days |
| Cosmos Hub | ATOM inflation (10% a year in its parameters) plus fees | Yes: 5% for double-signing with permanent removal; 0.01% for downtime with a 10-minute jail | Yes, pro rata | 21 days |
| Polkadot | DOT inflation | Yes for validators, per offence | No for nominators since July 2026 | About 2 days for nominators; validators keep their previous unbonding period |
| Solana | SOL inflation (3.57% a year on 8 Nov 2026) plus fees | No slashing active | No | Stake deactivates over about one epoch (around 2 days) |
| Cardano | Fees plus expansion of the ADA reserve | None by design | No | None: ADA stays in the wallet |
Cosmos Hub: 5% for double-signing, and delegators share it
The live parameters, read on 8 November 2026, set a 5% slash for double-signing and 0.01% for downtime. A validator is counted as down when it signs fewer than 5% of the last 10,000 blocks, and it is then jailed for 600 seconds before it can send an unjail transaction. Double-signing is far worse: the validator is also “tombstoned” and, in the words of the Cosmos SDK documentation, “can not rejoin the validator set”. Delegators are slashed with their validator, and the staking module documentation extends the cut to stake that began unbonding or redelegating after the fault; the slash applies when the evidence lands on chain, not retroactively. Unbonding takes 21 days and the active set is capped at 180 validators.
Polkadot: validators slashed, nominators spared since July 2026
Validators who sign two conflicting blocks lose a share that grows with the square of the number of offenders: min((3x/n)2, 1), where x is the number of offenders and n the size of the active set, according to the Polkadot wiki on offences. With 100 validators, one offender loses 0.09%, five lose 2.25% each, twenty lose 36%. Backing an invalid parachain block costs 100%, and slashes are held for 27 days so that governance can cancel them. The big change of 2026 concerns holders: referendum 1910, enacted on 6 July 2026 according to a Polkadot forum post of that day, made nominators no longer slashable and cut their unbonding period from 28 days to about two days (two eras). Validators remain slashable, with their unbonding and deferral periods unchanged. Pages of the wiki that still describe nominator slashing predate that vote.
Solana: no slashing active, two proposals on file
As of 8 November 2026 the Solana protocol has no active slashing. Inflation, read from a public mainnet node that morning, stood at 3.57% a year, all of it paid to validators and their delegators. Proposals exist: SIMD-0204 would add an on-chain program that verifies and records slashable offences, and states that it does not modify any stake or reward; it is still marked “Review”. A separate slashing proposal, SIMD-0212, was filed in December 2024 and its pull request was closed in January 2026 without being merged. What happens to Solana delegators when a validator fails was visible in October 2026: after a security incident at a staking provider, the Solflare validator stopped producing rewards, but the SOL delegated to it stayed in each holder’s stake account, as reported in our coverage of the Solflare validator incident.
Cardano: no slashing by design
The official Cardano glossary is plain about it: there is no lock-up, no slashing and no minimum amount, and the ADA stays in the holder’s wallet. A pool that performs badly simply earns less. Pool rewards are split by a fixed rule: the operator’s declared cost and margin first, the rest to delegators in proportion to their stake; a pool whose owners do not meet their declared pledge earns nothing for that epoch.
Bitcoin has no validators and no slashing: a miner who builds on the wrong chain loses the electricity spent, which is its own kind of penalty, as explained in our lesson on what consensus is. The next lesson of this path, delegated proof of stake, looks at how chains like Tron and the Cosmos Hub choose who validates.
How to check a validator before trusting it
Whether a holder runs a node or picks an operator, the same public data answers most questions. On Ethereum, explorers such as beaconcha.in show each validator by index or public key: its balance over time, the share of attestations included on time, missed proposals, and a slashed flag. A validator that is offline shows a run of missed attestations and a falling balance; a slashed one shows the slashed flag and an exit epoch.
For an operator or a pool, the useful questions are concrete. Does it publish the list of its validator keys so that their record can be checked? Has any of them been slashed, and what was paid back? Does its terms of service say who bears a slashing loss, and is there a bond or insurance behind that promise? Queue data, such as the entry and exit waits on validatorqueue.com, tells a new staker how long the money will sit idle before earning and how long it will take to get out. Our coverage of the Ethereum staking queue in October 2026 shows how fast those waits can change.
On Cosmos chains, the jail and tombstone status of a validator is public in any explorer, and on Polkadot the list of active validators and their nominators is on chain. A practice run with a tiny amount, or on a test network, costs little and shows how delegation, rewards and unbonding work before larger sums are involved.
Staking rewards and tax: income first, slashing unclear
The tax rules for rewards were set out in the previous lesson: in the United States, Revenue Ruling 2023-14 treats rewards as income when the taxpayer gains dominion and control over them, and in the United Kingdom HMRC’s Cryptoassets Manual (CRYPTO21200) taxes non-trade staking as miscellaneous income. The IRS ruling says nothing about slashing, so how a penalty that destroys part of the stake is treated is a question for a tax adviser rather than for a general rule. Reporting and later sales are covered in our guides to crypto taxes in the United States and in the United Kingdom.
What could change in the rules
Every number in this lesson moves. The consensus yield of 2.5% follows the square root of the stake and will fall if more ETH is staked. The penalty constants have changed at almost every Ethereum upgrade, and the next fork, known as Glamsterdam, was not scheduled on mainnet as of 8 November 2026 according to the live node configuration. On Solana, SIMD-0204 only records offences and the slashing proposal SIMD-0212 was closed in January 2026, so no penalty is scheduled. On Polkadot, the July 2026 vote moved the risk from nominators to validators, and wiki pages written before it still describe the old rule. The next Ethereum fork, Glamsterdam, has no mainnet epoch yet; until one is set, the constants quoted here, 1/4096, a multiplier of 3 and 8,192 epochs, are the ones in force.
