Does the world need software that recognizes AI in videos?

As you may know, a large amount of video content today is generated using artificial intelligence — with tools like Sora, Veo, Kling, and others.

The quality of this kind of video is constantly improving and has already reached a very respectable level. At the same time, it is extremely inexpensive compared to traditional video production with actors, camera crews, and everything that comes with it. I’m confident that in a year or two, about 95% of the video published on the internet will be synthetic — fully generated by AI. It feels like the world is heading toward a problem of widespread distrust in video content. We used to say that seeing is believing, but it seems that this rule no longer applies.

Our ML team has built a test prototype that detects signs of AI generation in synthetic videos with 99% accuracy. As mentioned, this is still just an experiment. It works very slowly — processing a 10-second video currently takes about 10 minutes. We can speed this up, but doing so would require additional resources, which is why I’m asking this question here.

Do you think a tool like this is actually needed?

Are there already similar solutions that work well, have proven themselves, and would be difficult for us to compete with?

Should we continue developing in this direction, or is this ultimately a pointless idea that nobody really needs?