"AI" things are indeed probably going to improve code security.

The irony is that, if so, they'll do it by increasing the amount of barely working, security-flaw ridden junk that gets put in use so drastically the problem finally becomes unignorable at a societal/policy level.