I feel like googling in itself is not a bad addition. It can help even the playing field, and it also disincentives to set "knowledge checks" as problems. It doesn't really help with the problem solving process, but it can help fill knowledge gaps / find programming templates you currently don't have on you.

It mainly feels unfair if it is ridiculously easy to google the entire solution to a problem: Div1F I think for div1F just typing in the keywords in the statement in any sane search engine would lead you to the solution. (My path was: go to the stack exchange post, then see that there's a paper, search up the paper and try to find an easy to access pdf.) I spent most of the time in contest fumbling with template code for div1F I found on the internet.

The main issue here is in contest preparation. When you set problems on codeforces you should pre-emptively try a lot of search queries into regular search engines and OEIS, just to make sure a problem is not well-known. Also having testers from diverse backgrounds can help against it. I feel like the necessary precautions were not taken this time.

If I reduce a problem to a problem I can reasonably be sure is classical, why should I not be able to solve the problem just because I haven't already prepared the solution in advance? E.g. for atcoder ABC456E, I reduced the problem to detecting a cycle in a directed graph. I didn't immediately know how to solve this reduced problem, maybe there is some value in figuring it out by myself, but I don't think it's interesting, so I just copied it from cp-algorithms.com.

I would go even further: in an ideal world, we would be able to use less powerful LLMs that can't instantly solve the problem, but can still be used for doing tedious but trivial optimisations. E.g. for ARC224E, I spent >10 minutes replacing a vector with a linked list, and ran out of time. (Granted, my solution would have failed regardless; I got ChatGPT to do it after the contest, but I TLEd for one less test case, and I potentially could have got it eventually.) The only reason all LLMs should be banned is because it's very difficult to create and enforce a rule that says that some LLMs are okay, but others are not.

honestly just do what is best for you. You can google the problems, copy a problem into GPT, but why? It isn't going to make you better.

The only problem with this approach is contests with prizes. Most AIs are still worse than top 100, so its fine I think? I am definitely not an expert though.