News for hackers. Stuff that matters.
Re:Ethereum Services Are Centralized by a1369209993 on Tuesday October 12, 2021 @07:41PM (#28846471)
> to deal with cryptographic breaks in its chosen cipher suite

Interestingly, this is actually almost doable in the context of a decentralized contract/computing system (with bug bounties to incentivise reporting[1]). You can include in the contract a interface where someone provides a proof[0] that a given cryptographic primitive is broken (eg a pair of colliding messages for a hash function), and the contract self-modifies to no longer trust that hash function.

The other half of the process - adding new primitives - is harder, since you want to prevent the addition of deliberately broken primitives. Obvious ideas include requiring a machine-checkable proof of hardness from some reasonable axioms (eg ZFC + P!=NP + whatever else), although most primitives don't have such proofs, and that doesn't look likely to change. Or requiring a quorum of reliable-seeming cryptographers to vouch for it (refutation: NIST endorsed Dual_EC_DRBG). A escrow period, where a new primitive is accompanied by a deposit that's paid out to anyone who breaks it, might help, but you'd need a seperate mechanism to prevent NOBUS backdoors (again, see Dual_EC_DRBG).

0: Ideally, you want a semi-zero-knowledge proof such that someone who has the option of putting in a tractable but significant amount of work to prove it (eg hash collision in 2^64 operations) will expect to be able to recoup their losses without getting sniped by front-runners, but once such a proof has been used, it's only a matter of time before front runners recover it and warn every contract.

For hash collisions, probably the easiest design is a two-phase model, where the reporter provides a zero-knowledge proof to commit to a particular hash collision, but then has to provide the actual collision to get the bounty. This can also work for preimages of a Schelling point like all-bits-zero. I'm not sure what a good Schelling point for typical public-key cryptography like curve25519 would be though; zero is almost always algebraicly special.

1: Cryptographic primitives are generally broken incrementally, so this incentivises even malicious parties to report a not-yet-exploitable break immediately (and thereby neuter it), rather than spend time trying to develop a working exploit (and risk both the large and small rewards disappearing out from under them because someone else was either honest or just less patient).

There are no comments on this comment yet.