Given how many pictures governments and corporations collect from public places, the GP's concern seems moot. I'll try to articulate my reasons as follows:
- In every authentication system (the airports' face scanning ones and others) there's a point at which a yes/no decision must be made: is this person authentic or is not?
- This yes/no "decision module" must base its determination solely on a series of bits presented to it by the image sensor.
- Every series of bits can be spoofed because the decision module can't tell whether the bits originated from a real image sensor or from a very convincing AI or elsewhere. The only exception to this is when the bits include a cryptographic signature, generated using a private key, securely embedded within the image sensor.
- The chance of such spoofing is minuscule if the sensor and the decision module coexist within a single piece of hardware that's tamper-proof. The decision module for airport face scanners can't be, given the large number of faces that must be queried. When such a decision module and its image sensor are separated by a network, possibilities for intrusion and spoofing can no longer be ignored.
- A helpful analogy is how we decry passwords stored as plain text in backend databases; after the inevitable compromise, these passwords get used to attack other systems. If backend systems store face data as a set of images (as I believe most do), how's that different in principle from storing passwords in a DB, in plain text?
- I'll grant that a careful designed system will allay my fears. The backend should store nothing but salted hashes and the image sensors must send only signed images of the subject.
- Stepping back, my ultimate concern with face authentication systems is that their technical details are opaque and they're used in situations where recourse is limited at best.