Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Similarity attacks can be mitigated in the following ways:

[1] Create as much perceptible variation as possible - TV static and minor color changes are not. [2] Make representations humans can remember - faces or images over abstract objects/shape color combinations [3] Make generating each representation expensive, so that searching for a visually similar fingerprint is expensive.

IMO most hash visualizers fail at all 3.

[3] is similar to hashing itself, except "visually similar" is substituted for "exactly equal". This is already taken care of by the hashing algorithm if it is slow, e.g. password hashing algorithms.

[1,2,3] could be solved with a GAN. Note that the GAN does not have to worry about making similar hashes produce dissimilar outputs, because finding a similar hash is already hard.



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: