Re:/dev/random – a new approach
by lvs on Sunday April 24, 2016 @05:37PM
(#11561478)
A malicious module would only need to act like VW's defeat device to overcome that check. When it senses some integrity/signing routine, it simply isolates itself from a verifiable circuit.
[ Parent
| on HN ]
There is no procedure you can run that would show that the RNG was compromised. They would still give you a stream that's indistinguishable from random noise, except it would be generated from a seed known to the attacker.
For VW, that was easy because the test setup was known and very rigid.
If you randomly check, say, power usage and latency of the on-board RNG (ideally combined with high-resolution IR recordings of the CPU), it will, at least, be harder for the CPU manufacturer to install a backdoored RNG that is active most of the time (but one that only gets activated after receiving some trigger, say a specific instruction sequence would still be virtually undetectable)