It depends a lot on your priors. I think making existing cryptography researchers hyperefficient and much more mathematically capable is likely to generate some disruptive results, but you may disagree.
Don't forget LLMs can now design efficient high speed hardware systems, a traditional slow-down for many attackers which represents a significant barrier to scalable next-gen attacks falling. Unfortunately memory prices are through the roof, but that isn't always significant.
LLMs probably won't break curve cryptography. Cryptography researchers armed with LLMs are a different story.
Okay and why should we believe that?
It depends a lot on your priors. I think making existing cryptography researchers hyperefficient and much more mathematically capable is likely to generate some disruptive results, but you may disagree.
Don't forget LLMs can now design efficient high speed hardware systems, a traditional slow-down for many attackers which represents a significant barrier to scalable next-gen attacks falling. Unfortunately memory prices are through the roof, but that isn't always significant.
Your probability distribution should be flatter at least.
It's a risk
You don't