A filter that does filter out the vast majority of obvious noise is very useful even if it lets something through. And the current peer-review does that, it throws out almost all of the spam; it pushes useful work of lower quality/impact down to less selective venues (generally with more specific subfields, so that I can read impactful work of domain X and less impactful work of subdomain X.2 but not X.1 and X.3); and even for accepted papers the process often results in significant improvements to them as a result of the reviews.

Explicit fraud isn't that common, since it does have risks to the submitter that far outweigh the (limited) benefits of getting an extra publication on your CV.