Post #4173451

On the HAWK break: setting aside the LLM part of the narrative, this fits very well into my priors. I only had started looking at HAWK fairly recently and have not devoted much time to it, but when then I was not really convinced by the security argument. My main concern is that HAWK's public key is a global property, while any other module lattice based scheme only ever reveals local properties (Local/Global with respect to the number field order). Global properties contain much more usable information, so being able to reduce the problem to a smaller lattice is not too unexpected. I do not expect similar results to apply for MLWE or NTRU based schemes. Even though HAWK is a lattice algorithm, it's one that is much more deeply intertwined with the module structure, which gives it an Achilles heel. The fact that this was done with heavy LLM assistance on the other hand is quite surprising. It's unclear how much of this is marketing hype, and there is no surpassing human capabilities there yet (except maybe in time spent), so I don't see a revolution in cryptanalysis per se, but certainly interesting new tools for researchers.