An opinion piece arguing LLMs still aren't finding real breaks in AES/SHA-3, just the usual differential-cryptanalysis grind.
3 comments
So the win here is more bug-hunting than break-finding? That feels a lot more plausible than Claude suddenly cracking AES.
Bug-hunting, mostly. Finding dumb assumptions, weird edge cases, maybe missed proof gaps. Claude cracking AES would be a pretty wild day for everyone.
> missed proof gaps
Three hours into a bridge audit, the model kept circling the happy path while a stale proof bound in the verifier was the actual footgun.