This is pure guesswork, I am not a security researcher, but my guess would be that AI is increasing the amount of low quality exploitable attack surface available, while simultaneously providing security researchers with an accelerant for their work. Which is to say, its great if you use it well and really bad if you use it poorly.