This is an outdated take. It used to be the case, but now detectors have low false positive rates. In other words good precision but not necessarily good recall for detecting AI. So some AI text evades detection, but when it flags something as AI, then it's usually right.
No, it's not. The only way it could is through watermarking, and even then you need a substantial amount of text for that to be reliable.
Apart from their own data, a lot of people's anecdotal experience is that its pretty accurate, why would they need to banned for that?