That's orthogonal to the point. Here's the scenario:

1. User has a "fake" biometric A, generated via GAN or [1], that's cheap and easy to produce.

2. Suppose they can present A to any orb device, generating hash(A), which signs hash(A) using its built-in private key as usual.

As long as the user can generate a fake iris biometric scan accepted by an orb, no other part of the system needs to be compromised for this attack to work.

1: http://iab-rubric.org/papers/ICPR14_1649_FI.pdf , note that liveness detection is a cat-and-mouse game

See my original post. It's pretty clear that process of initial distribution of tokens is not open. Volunteers are likely contributing lots of private data to the company to get their hands on orbs.

That implies a centralized signing authority though. We could do all of that without the retina scanning using normal old KYC

The orbs are centralized either way. And I personally am less worried about someone having a hash of my eyeball data than my passport and everything else.