When it comes to mathematics, I fully agree with another roof here. We do mathematics because of its beauty. Just like slop images are not art, slop proofs are not mathematics.
https://youtu.be/o0MjPe7wXs8?is=iGMIPGk71yu89cCC
Unfortunately, when it comes to cryptography, I cannot be as principled. An attack is an attack and requires analysis and mitigation, even if it is ugly as the night.
That being said, so far AI has not produced any cryptographic results that are outside of my prior expectations. Mostly it just does the druggy technical grunt work that humans find too boring and too little pay off to do.
E.g. turning an attack on RSA requiring raw signing oracles into prehashing signing oracles. Not surprising, but technical and kind of boring. (And yeah, that RSA paper doesn't really have much practical applications, aside from maybe blind signatures, whether with raw signing oracle or with PKCS1 prehashing signing oracles)