You're making the unreasonable assumption of good faith on the part of the app developers.

It could be designed to accurately distinguish citizens from noncitizens, or it could be connected to a database of online agitators, or picking out facial features of targeted minorities without regard to their personal identity, or some combination of all of these and worse. You don't know.

Faces can’t be easily be “faked” in person but they can definitely be misrecognized or recognized correctly against mistagged data. Anything like this needs to be designed with the assumption that errors will be frequent enough that the finding has to be validated before doing something serious.