Seeing how some people talk about AI, I would expect there would be major security breach every day nowadays. Not one. Several of them. Where are they?
Some people would say, that the edge models are so protected, that bad actors cannot use them. But even models from few months ago were claimed to be able to find holes anywhere. And there are chinese and other models that do not have such limitations.
So either the models are not good at actually breaching the systems, but might find some holes. Or the systems are so well designed that they are hard to breach despite. Or most system are simply not worth hacking in. Or just having the exploit does not mean that the people driving an attack have the will or courage to use it.
Anyway, I would expect much more severe breaches to be happening if LLMs are actually as good as claimed, so I think the claims are overstated.