With a theoretical Casper liveness denial attack, the cartel is not trying to revert blocks. Instead, a cartel of 34% or more of validators would simply refuse to finalize any more blocks. Casper’s BFT-style consensus blockchain would still grow, but with a much lower level of security.
If no blocks are finalized for some long period of time, like a full day, there are two options for Casper PoS to defeat the attack:
1. Automatically rotate the validator set. Blocks under the new validator set finalize, but clients know new finalized blocks are suspect, as it’s very possible old validator set will resume operating and finalize some other blocks. Clients then manually override this warning once it’s clear that the old validator set is not coming back online. All old validators that did not try to participate in the consensus process suffer a large penalty to deposits.
2. An Ethereum hard fork adds new validators and deletes attackers’ balances. The new fork is coordinated one or two ways:
– Market consensus. Old and new validator set branches would both trade on exchanges. The market will “want the good guys to win” and the chain with validators that have demonstrated their goodwill would overtake the other chain.
– Social consensus
There is a spectrum of response strategies between in-protocol automation and social coordination, but automated resolution is preferred to minimize the risk of a simultaneous 51% attack and attacks on market consensus and/or social layer.