> A live human detective confirmed it.

I doubt it, due to human nature. Perhaps the process says the human must consciously validate, but a lot of humans in many cases will just rubberstamp what the AI said. That's the risk.