Slashing & Validator Economics

The three slashable offences

Slashing is defined by signatures, not intentions. Anyone who sees two conflicting signed messages from one validator can submit them as evidence, and a block proposer puts that evidence on-chain. There are three offences399:

• Double proposal: signing two different blocks for the same slot.
• Double vote: signing two different attestations for the same target epoch, in effect backing two competing chains in one round.
• Surround vote: signing an attestation whose source-to-target span strictly contains that of another (source₁ < source₂ and target₂ < target₁). This is the vote pattern needed to finalise a conflicting history.

These rules give Casper FFG its safety guarantee. Two conflicting checkpoints can only both be finalised if at least a third of the stake breaks one of them, and each of those validators can then be proven guilty.

Being offline is not slashing

A validator that is switched off signs nothing, so it cannot contradict itself. It is penalised under separate, milder rules.

Normal times. Since Altair, attestation rewards are weighted for timely source (14/64), target (26/64) and head (14/64) votes. A validator that misses an attestation is fined the source and target shares, and nothing for head400. For a 32 ETH validator at current stake levels that is about 0.0014 ETH a day, a little less than it would have earned. A week offline costs roughly a week’s rewards.

The inactivity leak. If the chain goes more than four epochs without finalising, usually because more than a third of stake is not voting, an emergency mode starts401. Each offline validator gains 4 “inactivity score” points per missed epoch, and its penalty each epoch is proportional to its score, so losses grow quadratically400. Online validators earn no attestation rewards during a leak but are not fined. The leak drains offline stake until online validators again hold two-thirds and finality resumes. A 32 ETH validator offline for the whole leak would reach the 16 ETH ejection balance in roughly three weeks and be removed402.

What a slashing costs

When evidence is included in a block, the protocol imposes four costs403:

1. Initial penalty: 1/4,096 of the validator’s effective balance, burned immediately. That is 0.0078 ETH for a 32 ETH validator, or 0.5 ETH at the 2,048 ETH maximum.
2. Forced exit: the validator joins the exit queue and can never rejoin. Its balance cannot be withdrawn for at least 8,192 epochs (about 36 days), and longer if the exit queue is long.
3. Missed-duty penalties: for those 36 days it cannot attest but is still fined every epoch for missing source and target votes399400.
4. Correlation penalty: about 18 days in, the protocol totals the effective balance of every validator slashed in the last 8,192 epochs, a ~36-day window centred on this slashing. It triples that total and caps it at the total stake. The validator then loses that fraction of its own effective balance403404. If a third or more of all stake was slashed in the window, it loses everything.

The proposer that includes the evidence earns a whistleblower reward of 1/4,096 of the slashed validator’s effective balance.

What Pectra changed

Pectra (May 7, 2025) included EIP-7251, which raised the maximum effective balance to 2,048 ETH so that large operators could consolidate many 32 ETH validators into a few405. Under the old 1/32 initial penalty, slashing a 2,048 ETH validator would have cost 64 ETH at once. The EIP therefore lists “making the initial slashing penalty negligible” as one of its defining features405. The penalty constants have changed with each upgrade399400404403:

Electra also stopped rounding the correlation penalty down to whole ETH. Before, an isolated slashing usually paid zero; now it pays a tiny amount403. The forced exit and the 36-day lock-up did not change.

A worked example

Take about 43 million ETH of active stake, roughly the September 2026 level406, and a validator that would otherwise earn the full consensus reward. These are approximate figures from the spec formulas.

A 32 ETH validator slashed alone loses 0.0078 ETH in initial penalty, about 0.05 ETH in missed-duty fines and a correlation penalty of 3 × 32 ÷ 43,000,000 × 32 ≈ 0.00007 ETH. The total is about 0.06 ETH (0.2%), plus about 0.08 ETH of rewards it would have earned over those five weeks.

The same validator in a correlated event:

A consolidated 2,048 ETH validator slashed alone loses about 4 ETH, also about 0.2%. That is 0.5 ETH initial, ~3.2 ETH in missed duties and ~0.3 ETH of correlation penalty, which is larger than a small validator’s because its own balance counts towards the window. Before Pectra, 64 separate 32 ETH validators slashed together would have paid 64 ETH in initial penalties alone.

That is the design: an honest one-off mistake costs little, while an attack, which needs a third of the stake, costs every participant everything.

Real slashings and what caused them

Each well-documented slashing comes down to the same failure: a key signed the same duty twice, because two copies were running or because the record of past signatures was lost.

• Staked, February 2021. 75 validators were slashed together, about 18 ETH in total under Phase 0’s smaller penalties. Chasing performance, the operator had stopped persisting the Prysm client’s slashing-protection database. Its validators restarted more often than in testing and signed duties they had already performed. Customers were compensated, and the firm said the gains “weren’t worth the additional risk”407.
• Launchnodes (Lido), October 2023. During a datacenter failover, the original validator client and a manually started fallback were both connected to one remote signer that had slashing protection turned off. 20 validators double-voted. Penalties and missed rewards totalled about 28.7 ETH, including 20 ETH of initial penalties under the old 1/32 rule. The operator compensated stETH holders408.
• SSV Network, September 2025. 39 validators were slashed, the largest correlated event since Pectra. SSV’s post-mortem says an operator’s maintenance mistake ran a parallel validator instance outside the distributed-validator cluster, and that the DVT protocol was not at fault409. For a 32 ETH validator, the initial penalty under the Electra rules was under 0.01 ETH.

None was an attack. The backup meant to improve uptime is what signed the second message.

How operators avoid being slashed

• One key, one active signer. Start a fallback only when the primary is provably dead and its keys are gone, not just when a management API says they are.
• Keep the slashing-protection database. Validator clients record what they have signed and refuse to sign anything that conflicts. EIP-3076 defines a standard export format, so a key moved between clients or machines carries its history with it410. Remote signers need this protection turned on at the signer.
• Doppelganger detection. Many clients can watch the network for a few epochs after starting, to check that the key isn’t already active somewhere else.
• Distributed validator technology. Obol, SSV and similar systems split a key into shares across several machines or operators, and a threshold of them (for example 3 of 4) is needed to sign. One failure doesn’t take the validator offline, and no single machine holds the whole key402. DVT replaces the risky hot spare, but it cannot protect a key that is also being run outside the cluster.
• Client diversity. Penalties grow with correlation. A bug in a client run by over a third of stake can trigger an inactivity leak. A client run by over two-thirds could finalise a faulty chain, leaving its users heavily penalised.

Other proof-of-stake chains

• Cosmos Hub. Double-signing burns 5% of the stake bonded to the validator, delegators’ stake included, and removes the validator permanently. Missing more than 95% of the last 10,000 blocks burns 0.01% and jails the validator until it unjails itself. Unbonding takes three weeks, and stake can still be slashed during that time for earlier faults411. Downtime costs principal here, not just rewards, but there is no correlation penalty.
• Cardano. There is no slashing. Delegated ADA never leaves the holder’s wallet, and a poorly performing pool simply earns less412.

Ethereum is lenient on downtime, mild on isolated equivocation and severe on coordinated equivocation, the only kind that threatens finality.

The economics of running a validator

Stake. A validator needs 32 ETH to activate. Since Pectra, one with compounding (0x02) withdrawal credentials can grow to a 2,048 ETH effective balance, so its rewards compound. Older 0x01 validators stay capped at 32 ETH, and any excess is swept to the withdrawal address405402.

Consensus rewards fall as stake rises. Base rewards are proportional to effective balance divided by the square root of total stake400. Total issuance therefore grows with √stake, and the yield per ETH falls as 1/√stake. Ethereum chose the square root on purpose: it keeps “discouragement attacks”, in which a validator profits by driving others away, unprofitable413. With perfect performance, consensus APR ≈ 166 ÷ √(total ETH staked):

Execution-layer rewards are lumpy. A block proposer also collects priority fees and, through MEV-Boost, a builder’s bid for the block (see MEV). Proposers are chosen at random, weighted by effective balance403. At about 43 million ETH staked, a 32 ETH validator proposes about twice a year on average, and one proposal can coincide with a large MEV opportunity. Pools spread this luck across thousands of validators. A solo staker can’t.

Costs. ethereum.org recommends a 4 TB NVMe SSD, 64 GB of RAM, a modern multi-core CPU and about 50/25 Mbps of bandwidth, online all the time402. On top of that come electricity, your own time and the opportunity cost of locked ETH. Entry and exit go through rate-limited queues, and in early September 2026 the entry queue was over a month long406.

What it pays. On September 7, 2026, with about 42.9 million ETH (35.6% of supply) staked, one large operator reported a 2.62% APR, and Lido’s stETH showed a 7-day APR of 2.60% after its fee406. A fair range as of late 2026 is about 2.5–3% before fees, with tips and MEV driving most of the week-to-week variation. These figures are in ETH, not dollars. Part of the yield is paid for by issuance: about 1.1 million new ETH a year at this stake level, before fee burning, which dilutes holders who don’t stake413.

Solo, pooled or liquid

Solo staking keeps every reward but leaves all the operational risk with you. Liquid staking hands that risk to operators and gives you a tradable LST, for a fee plus smart-contract and concentration risk. Operators compensated stakers in the incidents above by choice, not because the protocol requires it.

Key takeaways

• Ethereum slashes for three provable acts only: a double proposal, a double vote or a surround vote. Being offline is never slashing.
• Since Pectra, the initial penalty is 1/4,096 of effective balance. A slashed validator is still forced out, locked for about 36 days, and charged a correlation penalty of three times the share of stake slashed around the same time.
• A lone slashing costs about 0.2% of stake; a mass event can cost everything. What makes slashing expensive is correlation, not the act itself.
• Real slashings come from redundant setups signing twice. The defences are persistent slashing-protection data (EIP-3076), doppelganger checks and properly isolated DVT.
• Consensus yield scales with 1/√stake, about 2.5% at roughly 43 million ETH. With tips and MEV, staking paid about 2.5–3% before fees as of September 2026.