If you're talking about shirts with faces on them, those won't work whatsoever. Even if there is evidence that it appears to work right now, it's an entirely solvable problem today, with very little effort required.

a) You're assuming that the AI is looking to find the first face it sees, rather than all faces in view - both your shirt and your real face will be picked up as two separate individuals. Even if it's "one face at a time", why would you assume the shirt gets picked up instead of your face?

b) It really would not be difficult teach a neural net to detect one real face located above a face on a shirt, and ignore the lower one. The only potential false positives would happen with a taller individual walking with a shorter person in front of them, whereby the shorter person may be filtered out as a shirt.

So no, shirts with faces on them are not a countermeasure. You're adding additional noise, but you're not eliminating your own face from being picked up as well.

No. The idea is that with a need to conduct covert ops in a world filled by automated camera systems, there'd be a back-door built in, such that whenever one tries to retrieve footage of an "ugly shirt" (a special, machine-readable pattern) - that footage would be deleted.

That way a camera blackout/missing footage wouldn't signal a covert action, while at the same time one could go on defending democracy(tm) without worrying creating a media-storm about the tyrannical methods used to ensure Freedom(tm).