5 ms·
Forcing VW to release source code would not have prevented this. They could have simply released a different set code from what is actually running on the cars.
by NickM 11y ago
Forcing VW to release source code would not have prevented this. They could have simply released a different set code from what is actually running on the cars.
This is an extremely hard problem to work around. They could let you dump the binaries of the software running on an individual car, and then you could compile the source code and compare the resulting binaries, but how do you know the car isn't feeding you a fake binary dump? It seems like a catch-22: I can't think of any way around this problem short of tearing the car apart, cutting all the chips open, and physically verifying them under electron microscopes.
On the other hand, if emissions testing would actually test what's coming out of the tailpipe under normal driving conditions, then that would seem pretty foolproof.
- alephnil 11y agoI agree that actually measuring the real emission is a good idea, but there is a way of checking that the binary and the source does not match, at least in theory. The way is to require reproducible builds. If the source is built with a specific compiler with known source and known parameters, and the source of the system in question is available, then running the compiler will produce the exact same binary every time if the compiler is sane. If all of this is published, then it is possible to check that the binary in the cars firmware and the binary produced by the published source match bit by bit. This isn't just some sort of academic exercise, but something that have been demonstrated to work in practice. In fact debian try to make this work for the whole distribution. This may not be totally practical yet, but this kind of checking can be done if the vendors are required to make reproducible builds and the regulators bother to check. All of this are big ifs, but at least in theory it is possible.
- GhotiFish 11y agois there a way to guarantee the binary we get from the control system in the vehicle is the binary that is running on that vehicle? That was the concern NickM had for this style of verification.
- NickM 11y agoYes, exactly. Reproducible builds can catch accidental differences, but discrepancies introduced by a malicious actor can still easily be hidden if they have complete control over the software and hardware.
- ableal 11y ago> released a different set code I believe that, for instance, the GPL demands the toolchain be replicable. So under that constraint - really basic for any scientific purpose - the first test would be to run the proverbial 'make; make install' ...
- NickM 11y agoYou're assuming the car doesn't also cheat on the "make install" test. How do we know the car isn't programmed to secretly leave the cheating software intact? Whatever extra layers of software verification we attempt to throw at the problem can be defeated by extra layers of deception from the manufacturer.