[PoignardAzur]

Yeah, I'd say "whether humans are involved" isn't actually a good metric for whether abuse detection is fair / avoids screwing you over.

A better proxy would be "how much the company spends per user to detect false positives". Whether it's human oversight for each case, or engineering time spent fine-tuning algorithms to exclude known false positives, the more the company spends, the less it's going to screw you over.

(In practice, companies want to spend very little, which is why you get underpaid Mechanical-Turk workers and slapped-together detection systems.)