Oops, my bad for didn't read the whole article.
Still, the technology has its weaknesses. Kim said the system carries a hallucination rate of about 15 percent, including instances where it misidentifies an object as a person, which is why human judgment remains the final call.
The AI only flagged the people (or the objects it misidentified as people), but the human still decides whether those people are worth checking on. I think it still the human's fault if a lot of innocent people get harrassed by the police.

Don't worry, they treat it as a tool, like how people treat doorbell cameras with motion detection.