Home | Notifications | New Note | Local | Federated | Search | Logout

Note Detail

Reply to @feld@friedcheese.us
Q.U.I.N.N.@icedquinn@blob.cat (2026-09-24 09:03:35)
@feld @anemone @romin @sun i feel like something like heartbleed is a very different story to "auth=none."

the JWT style breaches are so absolutely catastrophically bad in that they are so well documented that literally just asking claude to look at your code would have probably gotten you slapped. it's a special kind of incompetence.
---Reply---
Q.U.I.N.N.@icedquinn@blob.cat (2026-09-24 09:06:15)
@feld @anemone @romin @sun that being said when i read about ECC crypto i think djb had a comment that a common newbie failure is not checking points on the curve first. the whole point of the curvesafe stuff was to find ones that were hard to scuff. since ECC contains many curves, and its easy to forget to check a point in a domain somewhere, and this is one of those pwns that can even just legitimately be a simple mistake at 4am.

the part thats bonkers is they're using frameworks that already do all this for you. :blobcatwaitwhat: they correctly didn't roll their own, but they still just. failed to bother to put the data in the saltpack box in the first place.

i dunno how you even use libsodium wrong like this. it's hard to do. its specifically designed so you can't fuck it up. :cirno_for_reals:

---Replies---

ロミンありん@romin@shitposter.world (2026-09-24 09:15:31)
@icedquinn @anemone @feld @sun
>a common newbie failure is not checking points on the curve first
because you don't know when an ec operation is gonna land you on infinity, you need to check at every step and bail, that's why he introduced the twisted edwards representation, all points land on the curve no matter what, no exceptions