I get paid nothing to do this - it's considered "service", i.e. it's rude to submit a lot of papers and not review any in turn. (it turns out there are a lot of rude people out there) In general no one in academic publishing gets paid anything, although an "area editor" who tries to convince people like me to review papers might get paid a bit if they work at a lower-quality for-profit journal. (other fields have high-quality for-profit journals, but not CS)
Some of the papers I review may be fraudulent. I have no way of figuring this out, it's not my job, and I don't have access to the information I'd need to determine whether they are.
The use of images in certain life sciences papers has made it much easier to detect a certain class of fraud, although even these checks would be difficult for an individual reviewer to perform. (the checks could be integrated with the plagiarism checker typically run on submissions before they are reviewed, and I think some journals are starting to do this)
In CS it would be much more difficult to detect fraud, because there's no equivalent to the standard and easily-compared western blot and photomicrograph images in life sciences papers. However I'll note that a lot of CS venues are starting to have an "artifact evaluation" phase, where authors of accepted papers submit their software and a team of students try to get it to work themselves. It's not mandatory, but most authors try for it - the main purpose is to encourage reproducible science, but it also creates an environment where fraud is more difficult.
(I'm only aware of conferences which have artifact evaluation - none of the journals I know have tried this)