Cybersecurity of vibe-coded products

I was invited to a live broadcast on Izvestia to talk about cyber threats and the use of AI in this context.

Here's a snippet from the broadcast where I discuss the vulnerability of products written with AI.

Overall, nothing new—this has been a topic of discussion for a long time, and cybersecurity is more in trend than ever. Besides the vulnerability of vibe-coded products, there's another interesting fact: cybercriminals can easily set up their own model on a Mac Mini, train it on created viruses, and this neural network now churns them out at incredible speed.

Also, it turns out there is a fully trained model specifically for creating viruses. I won't give the name so you don't accidentally break your own PC with it. That is, it's not a local model but a fully public large language model. It's just that, among other things, it was also trained on viruses.

It's a big topic, can't cover it in one post, but I think you get the gist. Besides, I'm not a cybersecurity expert. For now, enjoy the video)