> 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).