I know, but you need to think like someone trying to get around the limitation with the lowest effort possible.
They don't feed it a video clip. They hold the camera in front of a screen playing the video. Use a low-end phone with a blurry camera to increase your chances.
> They hold the camera in front of a screen playing the video.
Neural networks are very, very, very good at detecting this.
It's much easier than detecting "liveness" (for whatever definition of that term you subscribe to).