Ah yes the lie being fed to me, aka having seen a lot of the exploits, as well as their RCAs clearly pointing to human written code. Unless you suggest that these AI companies have discovered time travel and are sending back their LLMs in time a la Terminator, but instead of killing John Connor, they're just... injecting slop before LLMs were a thing. Which is obviously a ridiculous notion.

Yes, actual slop - code never reviewed by a human, having no actual maintainer, etc. - can have major security issues. Today, considerably less than, say, a year ago. Reality is, most slopware is either abandoned in no time, or, the less likely option, get traction, get more people working on it, and over time, become less sloppy as people take ownership of various parts.

Meanwhile, an overwhelming part of the IT industry now utilises AI with humans in the loop. We are actually in the phase of the "find out and oh fuck backtrack on it" part of the whole "let's fire engineers because a manager with AI can do the work of a department" idea backfired spectacularly.