How so? Has anyone ever pulled the RAM sticks out of their computer to check that the entropy pool looks like they expect it to? Did they then verify that the L3 cache on the CPU had the same contents? Or did they simply ask the suspected backdoor CPU if the very memory it was just suspected of tampering with was clean?
If you're going to open Pandora's box, I don't think you get to pick and choose what hypothetical backdoors to take out. It's particularly odd to only select the subset of backdoors that you can easily defend against.
Except that if the CPU completely masked its changes it would be no threat. The trick is getting the system to use the bad randomness. I find it unlikely none of the people debugging or running slightly different kernels or drivers or rootkits would not notice something. To exploit RDRAND you would not have to worry about what all the code in the system is doing (highly volatile over different configurations and versions) but you would just need to monitor a few select kernel symbols.
If you're going to open Pandora's box, I don't think you get to pick and choose what hypothetical backdoors to take out. It's particularly odd to only select the subset of backdoors that you can easily defend against.