It might be possible to automate it! Google keeps posting these incredible recognition results. Perhaps they can apply it to the video/audio looking for items it has trained it to be inappropriate?