Crypto glossary
Long-range attack
A long-range attack is a proof-of-stake threat in which old validator keys are used to build an alternative chain from far in the past and fool new nodes.
In proof of work, rewriting old history means redoing all the computation since that point, which is prohibitively expensive. In proof of stake, producing blocks only requires signatures. Validators who once held a large share of stake but have since withdrawn it can no longer be slashed, so their old keys cost nothing to misuse. An attacker who buys or steals enough of those keys can sign a competing chain starting from a past block, and it can look perfectly valid.
Nodes that have been online all along are not fooled, because they already know the real chain and refuse to revert finalized blocks. The danger is for nodes syncing from scratch or returning after a long absence, which see two chains and cannot tell from the data alone which one is real. That is why Ethereum and other proof-of-stake networks rely on weak subjectivity checkpoints.
Long-range attacks remain largely theoretical on major networks, since checkpoint sync and client defaults close the gap. They are often confused with short-range reorgs or 51% attacks on live consensus; the defining feature here is the reliance on keys whose stake has already left the system.
Example
Years after leaving a proof-of-stake network, a group of former validators sells its old keys; an attacker uses them to forge an alternative history, but nodes that sync from a trusted recent checkpoint ignore it.