There is a big grey area between "not 100% correct" and "wrong and harmful". Like a faulty computer program that produces correct results in 99.999% of the cases a wrong mathematical statement can still yield useful results in practice. You just don't want to run into those pesky edge cases where it indeed is wrong.

Having a database of all mathematical proofs where all of them are checked for validity, however, is a very useful tool to actually find those edge cases where a proof is wrong and give future mathematicians a solid foundation.

Using AI as an extension of human capabilities is a given to me. Like with heavy machinery, we don't want to do the heavy lifting. We want to steer it properly to gain most benefit.