It really wouldn't be that hard to compare what you received from the foundry with the masks you sent them. There are services that decap ICs and attempt to reverse engineer them. Here's the first hit I found on Google: http://www.intelligservices.com/Services.htm
You wouldn't need the reverse engineering. All you would need to do is to compare that all the metal layers (and perhaps poly) match what you sent. That's an almost trivial comparison. The hard part of reverse engineering is in figuring out what the circuitry actually does, and you already know that, since it's your chip!
It could be possible for a fab to alter diffusion layers to change the functionality of a chip. That would be harder to detect by services such as the one I mentioned. But it would be very hard, very time consuming, to attempt to hack in a backdoor by only messing with base layers, rather than messing with metal or with poly (where changes are easily observed).
If there were to be a backdoor anywhere in a design, it would be in IP you used on your chip, that you got from either your foundry or from a 3rd party IP supplier. It would be easy enough to hide all kinds of stuff there.
To put it another way, what are some of the details that the author(s) of this article got wrong?
Nearly everyone reading this on a laptop/desktop system is doing so on an Intel processor. Go ahead and attempt to enable remote access without a supported proc/chipset/bios and then come back and complain about this unexpected remote access.
There is a lot of misunderstanding about this technology in the security world which strikes me odd because it's well-documented and readily available for your own testing.
As for trust, first, what we generally know is that any complex system is likely to have flaws. So any remote access system is likely a back door - intentionally or not. A physical off switch would be prudent (arguably switching off wireless network card and not connecting an ethernet cable is one option: but then your options are: no networking or possible vulnerability).
It's not as if any networked machine is "perfectly secure" - but it'd be nice to be able to set the scope of exposure even more easily, and granularly.
Second, one thing is trusting Intel to not intentionally install a back-door - keeping in mind that includes any "golden key" for updates. And not leaving any holes in the interface they expose (ie:getting the feature right). That in itself is a pretty high bar, imnho. But another thing is trusting Intel not to build in secret backdoors - that is; if they say, no, there's no IME in product X, then trust them on that. That is to me a much more trivial thing to trust. I know there will be bugs in the software and hardware, but I'm not all that afraid of maliciousness.
If I was working for a Chinese intelligence service the threat model might have been different.
In sum, given that there's bound to be bugs, and might be well-intentioned software upgrade support - I would prefer if it was easier to buy high-end parts without stuff like IME (this also applies to AMD btw) - or at least that it was easy to remove a jumper or toggle a dip switch to turn the stuff completely off.
[ed: finally re: server Vs laptop: there are some things that doesn't matter as much if it's exposed on a server: got my encrypted backups? Encrypted email? No problem. Got my secret key from my laptop? that is a problem]
There is a common misunderstanding in these discussions that the ME features of the Intel processor (included in nearly all modern procs) allows for hardware level remote access, which isn't true.
[1] https://en.wikipedia.org/wiki/Intel_Active_Management_Techno... [2] https://en.wikipedia.org/wiki/Intel_vPro