I have some rough estimates for developing safe software, that is to say software that has been proven correct against a high-level specification:
at least 10x the effort, by developers at least 1.5-2x the skill/salary level
(eg. not the Tumblr staff)
This puts the cost of safe software at 15-20x the cost of regular software, which explains why it is virtually nonexistent in the commercial sphere: the cost of failure almost never justifies the expense required to avoid it, even more so when the cost of being late to market is considered.
Anyone interested in software safety will invariably get drawn into improving the formalisms and the tools used to work with them to bring this ratio down.
Oh hey, finally somebody talks about something I know something about!
Here’s the biggest thing I can say: For Gods sake do not make “safe software” a regulatory requirement. Somebody will start writing SOPs that follow Best Practices as understood by a non-coding QA and soon, it will be more important to be in Compliance than it is to be, you know, safe. Sure, one is supposed to correlate with the other but I have not, so far, been impressed. You see, you can measure Compliance. You cannot measure safety. So Compliance becomes the goal, and good software practices can be sacrificed along the way if it helps reach good regulatory Compliance.
Compliance. Not even once.