What is slashing?

A protocol penalty that destroys part of a validator's staked coins for provable misbehavior such as signing two conflicting blocks or going offline past a threshold.

It is what makes proof of stake economically secure: attacking the chain costs the attacker their own capital. Delegators who stake through a validator generally share the loss, so choosing an operator with a clean record and sensible infrastructure matters. Exchanges and staking providers differ in whether they cover slashing losses, and the answer belongs in their terms rather than their marketing. See proof of stake and validator.

Most people meet the word in the risk disclosure on a staking product. It is worth reading, because the two most common reasons a staked balance shrinks are not the same thing and are often described in the same paragraph.

How it works

Two different penalties get grouped under the label, and only one is slashing proper.

Slashing punishes provable, attributable equivocation: signing two conflicting blocks for the same slot, or two conflicting attestations. The offense is provable from the signatures alone, so anyone can submit the evidence and the protocol acts on it. On Ethereum the consensus specifications impose an initial penalty proportional to the validator's effective balance, force the validator to exit, and add a correlation penalty that scales with how many validators were slashed in the same period, so a coordinated failure across many validators costs far more per validator than an isolated one (source: the Ethereum consensus specifications).

Inactivity leak is the second mechanism and is not slashing. A validator that is simply offline misses rewards it would otherwise have earned and, when the chain is failing to finalize, loses balance gradually until finality returns. That is a bleed rather than a cut.

The design intent of the correlation penalty is worth noting, because it explains a real operational choice: running many validators on identical infrastructure with one client makes a single bug an expensive correlated event rather than a cheap isolated one. See client diversity.

Example

Illustrative arithmetic on round numbers. Suppose an operator running 100 validators of 32 units each, 3,200 units in total, suffers a configuration error that causes all 100 to double-sign in the same window. Under a correlated design the per-validator penalty is a multiple of the isolated one, so the loss is far more than 100 times the penalty a single accidental double-sign would have cost. The same 100 validators split across independent setups and clients would have failed in ones and twos rather than together. The exact fractions come from the chain's own specification, and the pattern is the point: correlation is what makes slashing expensive.

Why it matters when you buy

If you are buying an asset in order to stake it through an exchange, whose loss it is when something goes wrong is a term of the product, not a property of the chain. The yield pages show which venues offer staking on an asset and where it is permitted, and the staking guide covers what the products actually do.

Related terms

• proof of stake — the system slashing secures
• validator — the party at risk
• staking — the activity that exposes you to it
• unbonding period — the wait before stake is free again
• exit queue — the throttle on leaving
• client diversity — why correlated failure is costly

Questions

Can I be slashed for going offline?

On Ethereum, no. Downtime costs missed rewards and, during a finality failure, a gradual balance leak. Slashing proper requires provable equivocation. Other chains define downtime penalties differently, so check that chain's specification.

Do delegators lose money when a validator is slashed?

Usually yes, in proportion to their delegation, unless a provider has contracted to absorb it. This is the main reason the operator's track record matters when the yield figures look similar.

Does slashing affect liquid staking tokens?

It reduces the underlying stake, which is what the token represents, so the token's redemption value takes the loss. See liquid staking token.