If the software in power plants were treated like their hardware it wouldn't be a problem.
If that OS gets reused by many people so it gets vetted by time I think that would be best. Perhaps a very stripped down Linux or BSD could fill the role. There is absolutely no reason to have things network configuration wizards, wobbly windows or other fluff. I think there are already systems and tools to help someone construct a custom minimal Linux OS (Yocto) and perhaps people already use these to construct appliances for highly sensitive systems.
I don't think we are that far away, at least some industries are pretty well covered already.
So what... we switch to a stripped down version of Linux that won't ever have the kernel updated and will need to continue supporting version 1.0 of samba because of some arbitrary codepath to load new versions of firmware.
It's easy to sit back and cast judgement when you don't deal with that industry... but they aren't doing it this way because it's fun.
> It's easy to sit back and cast judgement when you don't deal with that industry
I work making special purpose appliances right now, well maybe not right this second, clearly I am posting on HN, but I am at my desk where I should be writing software for one of these appliances. I am only describing exactly what my team is doing. We have a certification process too. I know exactly how painful it is. Bitching about the current and acting like it is immutable law won't fix it. Deciding on what "fixed" looks like then trying to change conditions to get there is one possible way to fix it. There are surely others but throwing our hands up and giving while making excuses isn't one of them.
So someone who has a device in them and lives for another 30 years is forced to undergo surgery in order to update to a newer version? And you SERIOUSLY think that's going to fly?
I guess we can just agree to disagree.
But your notion of how implants work is interesting.
At present pacemaker batteries need to be replaced every 5 years (sometimes sooner if they stop responding by radio). Most pacemakers are only certified for 2 battery replacements. The situation you describe is actually much better than reality right now.
Did you know there is a count of people who die to pacemaker bugs each year? Did you know that most pacemaker recipients have no right to see the software or pay to have an expert review the software in their pacemaker? I don't know what you think is going on in medicine now, but there is much that is deeply fucked up with it.
Also your point is pretty ridiculous when you consider two points. We can do some updates to implants wirelessly and without surgery. Then consider that when regulations change there is often a subset of things that are grandfathered in from the old rules. There are still cars driving around with no seat belts and it is legal (in most places) because they were manufactured before seatbelt laws and maintained.
Will you build it with virtual memory? Will processes have separate memory maps? Will they communicate via shared blocks? What kind of protocol and connections will be used for peripherals?
I don't see that happening.
This will probably mean hospitals and the FDA get much stricter about what is allowed on medical devices. You cannot update because of risk: that means you can only have certified software. Microsoft will not like being told they are not allowed in hospitals and in turn will be more secure.
It is a race, and whoever is doing this is showing their cards early which gives everybody time to fix things. (I will not claim this there is enough time)
Easy practices like these will keep getting easier. Security is the default position of computers when they are not assembled haphazardly, because secure is a subset of correct. As we fix bugs we increase security. Combine this with half a clue about operational security and defending against "cyber-weapons" is something your organization does without noticing.
We are even now getting to the point were Journalist know to ask questions like "Did your organization fuzz this piece of software?" (Did you read the article yesterday about the windows defender bug). The market is applying more pressure on security and we will figure out how to make the market happy, because that is how it works.
It's easy to have a "secure track record" when you say no to 90% of the outside code contributions and support a minimal amount of hardware. There's basically 0 chance that OpenBSD would ever be the source of medical devices.
I guess you could suggest they just get universally redundant systems and take rolling outages, but I'm guessing nobody would be willing to pay for the cost of implementing such a scheme.