Rarely is this salary guaranteed. It often requires TAing to supplement research funding and graduate students are often required to be deeply involved (or even solely responsible) for sourcing their own research funding. Stipends are also often poverty wages or even lower when you consider that stipends are usually only for 3/4 of the year and you are usually expected to continue research during the summer.
I know some folks who did PhDs in non-engineering fields at state universities who had to TA every single semester, often TAing multiple classes, just to earn ~$20,000 per year.
This is true in computer science, but not in all sciences, and certainly not in the humanities. Getting a TA-ship is considered undesirable in CS, but getting a TA-ship as a history PhD is considered rare and special.
There are a lot of problems, though. Your funding isn't guaranteed; you can be basically kicked out of the program (because very few people can afford to self-fund) for reasons that have nothing to do with you or your performance, and if this happens in your 4th or 5th year, you're fucked.
The other ugly fact is that even if you're not paying tuition, your advisor is. This means there's less money to send you to conferences or fund his lab, and it means he's under tighter financial constraints than he really should be. Your advisor gets his budget docked $80,000, and only $20,000 goes to you (the rest, to tuition). That's a bit ridiculous, especially because you're no longer taking many classes or using university resources except to do work on their behalf. There's no good reason you can't be paid $50,000 and your advisor has $30,000 more to fund his lab.
I know some folks who did PhDs in non-engineering fields at state universities who had to TA every single semester, often TAing multiple classes, just to earn ~$20,000 per year.