The hardware required for AR glasses to do what they are advertised to do (eg: translate street signs, overlay advertisements while you're trying to eat, etc) is sufficient to identify anyone walking by*.
Not hard to write software to display on the wearer's HUD, for example, random photos and social media posts of whoever enters their field of view.
*providing the other things in my comment are in place.
They literally blocked a useful feature (recognizing people you have already met) and promoted a feature nobody wanted (posting pictures of people without consent).
most likely before such a threat model gets public exposure, you're going to already have legitimate implementations of this technology that will seem positive and benign.
for example, apple already does face recognition on your photos and then ties them contact names. now a gimmicky but totally futuristic use-case that you can demo in a commercial: you send your friend a contact from your address book, and in addition to phone numbers and emails it has a chunk of trained dataset for the contact's face. now you go meet this contact in person, and your ar "recognizes" the contact for you. the commercial will involve a meeting at a fancy Manhattan restaurant "oh hey you're Jeremy!" (you're welcome, nameless apple employee reading this post)
but see totally benign. by the time hackers get to it, it'll be ebil darknet russians breaking nice and pure future tech.