It's not obvious that a script got caught in a loop and decided to write up a $5000 bill overnight.
Even so, I don't think it is AWS's fault that buggy code was left to run unattended overnight in the example you give.
It is absolutely 100% Amazon's decision to give you little to no tooling to prevent this in many scenarios other than going to their billing department and beg for a correction.
I do not use any service that can proceed to bankrupt me for personal use for what could be considered a normal use case. At work I'd pass the issue off to my manager who'd pass it off to a rep and deal with it that way, but there is no way I want to take that responsibility on myself to correct.
That aside, I also understand that they can't actually put a hard cap, since usage is measured after the fact occurs, and by the time the system triggers a halt, more usage has likely occurred. At the very least, I'd be fine with paying $X + 1 when I set an auto-shutdown threshold to $X if that was what they had to do to implement it.
Outside of that quibble, I find the other complaints to be less reasonable.
Back to your analogy. You get charge for these other dimensions that vary when you rent a car:
- make, model
- miles driven
- toll usage
- driver’s age
- duration of rental
The more accurate analogy here for the GP is the person returned a rental car to the wrong lot and was charged additional days until they discovered their error.