11 ms·
The data wouldn't go through them, nor would they be responsible for auditing the algorithms themselves. The comparison with the FDA was exact: they would simpl
by derefr 12y ago
The data wouldn't go through them, nor would they be responsible for auditing the algorithms themselves. The comparison with the FDA was exact: they would simply require the company to execute a study proving (to peer review) the data-integrity of each change they were going to make.
The one interesting thing is that this would likely enforce an open-core-SOA software development model: companies would be incentivized to build a "trust kernel" of services that the government regs apply to, exposing an API with stringent access controls; and then a view layer that consumes that API, which can have whatever sloppy code they wish. The trust kernel would then have to be at least shared-source to enable the peer review necessary for study. (The company couldn't just pass the code around within a cabal of trusted peer companies, since those peers might be unfairly positively-biased.)
- gajomi 12y agoSorry, I tried to write what I imagined was a clarifying comment, at least along the lines of what I thought you were trying to say. But while I was typing it you (who I imagine is the real authority on your own opinion) did exactly the same, but better.
- convivialdingo 12y agoThat's fairly analogous to the current NIST regulations. Unfortunately, NIST has been dragging 140-3 in draft form on for years. 140-2 was written in the 1980's and reflects very badly on current hardware and software practices. Another area you could look into is Common Criteria. I find these certifications to be much more modern. I've taken products through both processes. If you're going for more than the basic levels they can be quite rigorous and thorough.