Proof of stake networks penalise certain validator behaviour by destroying part of the deposit. The penalty exists to solve a problem that has no equivalent in mining, where the cost of participation is external.
Signing costs nothing without a penalty
A miner extending two competing branches must split hardware between them, so supporting both is genuinely expensive.
A validator signing two competing branches expends almost no resources. Signatures are cheap, and producing a second one costs nothing meaningful.
Without a rule against it, the rational move would be to support every branch, since one of them will win and the validator is rewarded either way. Consensus would never settle.
Slashing supplies the missing cost
The protocol defines specific provable offences, chiefly signing two conflicting messages for the same slot or producing votes that surround an earlier one.
Anyone can submit the two contradictory signatures as evidence, and the chain itself verifies them and destroys part of the offender's stake.
Because both signatures are on chain and verifiable by any node, the judgement requires no human assessment and no trusted reporter.
Offences are contradictions, not failures
Being offline is not slashable. A validator that stops attesting simply misses rewards and gradually loses a small amount through inactivity penalties.
The distinction matters because downtime is often accidental, while a signed contradiction is evidence that two things claimed to be the same validator disagreed.
In practice, most real incidents are configuration errors, typically a validator key running on two machines simultaneously, each unaware of the other.
Correlated penalties target coordinated attacks
Many designs scale the penalty with how many validators committed the same offence in the same period.
An isolated mistake costs a modest fraction of the deposit, while an offence committed by a large share of the network at once can cost most of it.
This distinguishes accident from attack without needing to determine intent, since a coordinated attack necessarily involves many validators acting together and an operator error usually does not.
The consequences for delegated staking
Anyone staking through a provider is exposed to the operational quality of that provider, since the penalty falls on the stake regardless of who runs the machine.
Operators mitigate this with redundancy that avoids duplicate signing, hardware modules that refuse to sign conflicting messages, and careful key handling during migrations.
Concentration is the systemic version of the same risk. A large operator making one mistake triggers the correlated penalty for everyone staked with it, which is a reason for stake to be spread across independent operators.