I'm not saying this is what's happening here, but here's a playbook I could see behind it:

1. Automatic/ML/AI/buzzword filtering that is, in blunt terms, kinda crappy

2. Preemptively and diligently correct for any such crappy decisions that would result in PR nightmares or credible lawsuits

3. Apply some portion of your effort to reviewing some portion of the cases that appeal. Hope (nudge nudge, wink wink) that those cases that you couldn't review, because you simply don't have the resources to review everything in depth, are those of your "enemies." You're not suppressing them; you just didn't have the resources to help them.

Result: The system punishes your enemies by design and you and your enemies can just blame the AI and your front line support, which everyone already loves to do. People who are suspicious can be painted as nutjobs, because it's obviously just the unfortunate but classic case of the little guy against the giant indifferent bureaucracy.

Thus any sufficiently influential entity can wield Hanlon's Razor as a weapon. You don't event need to plan for this; it is bound to happen. Parts 1. and 2. happening are foregone conclusions. When the opportunity for part 3. arises, I think people in all levels of the organization would quickly and tacitly realize that uneven review is in their best interest, if they don't just do it automatically out of laziness or tribal thinking.