Claude found better attacks on HAWK and a toy AES, because apparently cryptanalysis now has a chatbot intern.
2 comments
Somewhere around 2019 I had a little SHA3-heavy prototype on a Ryzen box, and the last thing I wanted was to revisit the hash choice, so I kept the boring path and moved on. A month later a neat algebraic hash someone had pitched for the next rev got a handwavey break in review and I spent a weekend unwinding the whole thing from the sponge layer down, which is about the point I stopped being cute about it.
The part about not wanting to reopen the hash choice is fair, but the review break sounds like the real lesson, not a reason to trust the boring path by default.