It would have to get swiped on the way from the manufacturer to the OEM. Once the OEM has sent it out, it's protected against this exact kind of attack. And while it may make sense for interdiction of a single package to a known target, doing the same with an entire batch of chips seems prohibitively expensive.
>Given everything that has to be in place for vPro/the ME to work
Again, per Intel's documentation. For all anyone here knows, data exfil begins the moment a certain sequence of bits crosses the right registers - and it's not like this is beyond the capabilities of what ME lets you do.
There is no good reason that the entire subsystem can't be disabled by the user, permanently. But, come to find out about it, the chips are configured to shut the system down within 30 minutes if the ME firmware doesn't pass checksum.
That, in my mind, puts it uncomfortably close to malware territory. Every one of these concerns evaporate if the ME area could be wiped or dumped - it's not as if remote management is some secret competitive advantage.
The point I'm trying to convey here is that every piece of information available on this thing comes straight from the horse's mouth, and the horse is not necessarily a trustworthy actor.
...have to trust Intel's publicly available documentation. In other words, this has to be taken on pure faith that it does exactly what they say it does in exactly the way they say it does it, and no more.
The problem with ME is that it consists of unauditable code that could be doing literally anything on the computer, completely transparent to the user. Furthermore, even if there is nothing untoward happening when the chip leaves the fab (again, must be taken on faith), a documented threat is having hardware interdicted, modified, and sent on its way.