But theoretically, you could implement the protocol in a privacy-preserving manner where the only thing that needs to be saved, is the hash of the biometric data, not the biometric data itself.
So lets say that your face + fingerprint + iris each outputs a value. Concat those and hash them, and you have a unique value that can be reproduced elsewhere, without having to store anything else but the actual hash of the input.
Again, I'm not sure if this is what they are doing, but if that's how it works, they wouldn't actually need to gather any biometric data, after creating the hash it can be thrown away.