Grinding Attack
An attack in which a validator repeatedly tries different parameters or strategic choices to bias a supposedly random process, such as leader or committee selection, in their own favor.
A grinding attack occurs when a validator or block producer has some influence over an input that feeds into a protocol's randomness generation — such as RANDAO (see Kaavon's "RANDAO" glossary definition) — and repeatedly tries different valid choices for that input, computing the resulting outcome each time, until they find one that produces a favorable result, such as being selected as the next block proposer or landing on a committee they want. Rather than breaking the randomness outright, grinding exploits the attacker's ability to selectively choose among valid options rather than being forced into one predetermined contribution.
Protocols defend against grinding attacks by minimizing how much influence any single validator's contribution has over the final randomness output, combining many participants' contributions so that no one party's choice can meaningfully steer the result, and sometimes adding a delay between when a randomness-affecting action is taken and when its outcome becomes known, making it harder to profitably grind in real time. A successful grinding attack doesn't break consensus safety outright, but it can let an attacker unfairly increase their share of block proposals or committee assignments beyond their honest stake-weighted probability.
Examples
- A validator attempting a grinding attack might simulate many possible values for their randomness contribution before submitting, to see which one is most likely to result in them being selected as a future block proposer.
- Combining randomness contributions from many independent validators, rather than relying on any single one, is the main protocol-level defense against grinding attacks.