I haven't found anyone who has actually reversed the functionality to audit what it's doing, though.
There is no proof that they are
>hardly anybody bats an eye.
Because only a certain niche of paranoid people are willing to believe that companies are trying to make their systems intentionally insecure so that you can be personally hacked.
This is not unique to NSA. Security agencies within other governments play the same games with supply chains.
If you feel different that's fine but with security the old adage 'better safe than sorry' applies very much and given the data collected so far the evidence of 'some old leak' times n is rather more solid than your belief in the opposite. FWIW Intels claims that they would not cooperate with the NSA are worth absolutely nothing because the USG has the power to tell Intel to stay quiet about such a deal.
If CISCO could be influenced by the NSA then so can Intel, either through supply chain interdiction or through more direct means. But to categorically claim that this is an impossibility is nonsense, the relevant questions are (1) has it been done? and (2) how would we ever find out given the locked up status of the ME? and (3) who past Snowden would be brave enough to leak the evidence if it exists?
* By patching the ME firmware itself - see the me_cleaner project, and methods documented here: https://puri.sm/posts/deep-dive-into-intel-me-disablement/ . This is Pretty Reliable; the runtime code has been deleted from flash.
* By setting a bit in the flash configuration, assumed to be added for the US High Assurance program: https://github.com/corna/me_cleaner/wiki/HAP-AltMeDisable-bi... , https://www.ptsecurity.com/ww-en/analytics/disabling-intel-m... . This is Mostly Reliable; the mechanism has been fairly aggressively reverse engineered and was added for a program with strict requirements.
* By sending an HECI command that says "hey ME, turn off your runtime" https://review.coreboot.org/c/coreboot/+/52800 . This is Somewhat Reliable; the method is well understood and seems to work but I'm not sure someone has done a deep dive audit into whether it could be re-enabled somehow.