If a vulnerability is practically impossible for a human, but possible for AI, then it's not security through obscurity. Technology just advanced past a line where new things can be exploited.
There might also be a day in the future where our current encryption methods are broken - it doesn't mean they're insecure and vulnerable today.
None of these vulns was impossible for a human. There just wasn't that many humans clever enough to find them, and those that did we're busy not hacking random things. There isn't a single type of new exploit that AI has found that I know about.
We have a billon hacking experts now is the big change.